MIE1624 Group 5 Final Project

Group Members:

Hamed Zakeri, Farshad Tajeddini, Sina Davari, Gurkaran Sran, ​Samiramis Khazaei, Anbumanivel Mohan Suganthi, Dhilip Ilanseran​

Note: This notebook was run on Colab Pro because of higher RAM requirement. There is possibility of runtime crash if run with 12GB RAM which is available as standard.

Title: Sentiment Analysis on how Ukraine is perceived on the international scene during the Russian invasion of Ukraine

Problem Statement and Introduction:

The Russian invasion of Ukraine has been a widely discussed topic on the international scene. The way in which the conflict is perceived by the public can have a significant impact on foreign policy decisions and public opinion towards the involved countries. Therefore, sentiment analysis of news, tweets, and articles about the Ukrainian war can provide valuable insights into the attitudes and opinions of people towards this conflict. The aim of this project is to conduct sentiment analysis on a labeled dataset of text and gather our own cleaned dataset, train at least four classification algorithms, and compare the results obtained from a pre-trained model. Additionally, the project aims to determine the factors that influenced the sentiment derived from the analysis by using existing NLP models and visualizing the findings through techniques such as word clouds.

Source: https://www.everydayhealth.com/columns/health-answers/a-therapist-speaks-on-coping-with-feeling-anxious-sad-helpless-or-hopeless-about-ukraine/

# Import required libraries
! pip install langdetect 
import pandas as pd
import numpy as np
import nltk
import collections
import seaborn as sns
import matplotlib.pyplot as plt
nltk.download('stopwords')
nltk.download('wordnet')
nltk.download('punkt')
from nltk.corpus import stopwords
from langdetect import detect_langs
from nltk import word_tokenize
from nltk.stem import PorterStemmer
from sklearn.model_selection import train_test_split
from sklearn.feature_extraction.text import TfidfVectorizer
from sklearn.naive_bayes import MultinomialNB
from sklearn.linear_model import LogisticRegression
from sklearn.svm import LinearSVC
from sklearn.ensemble import RandomForestClassifier
from sklearn.model_selection import GridSearchCV
from sklearn.metrics import accuracy_score
from sklearn.metrics import f1_score
from sklearn.ensemble import VotingClassifier
from wordcloud import ImageColorGenerator
from bs4 import BeautifulSoup
from wordcloud import WordCloud
import datetime
import math
from sklearn.decomposition import LatentDirichletAllocation
from gensim import corpora
from gensim.models import LdaModel
!pip install pyLDAvis
import pyLDAvis.gensim

Looking in indexes: https://pypi.org/simple, https://us-python.pkg.dev/colab-wheels/public/simple/
Requirement already satisfied: langdetect in /usr/local/lib/python3.8/dist-packages (1.0.9)
Requirement already satisfied: six in /usr/local/lib/python3.8/dist-packages (from langdetect) (1.15.0)

[nltk_data] Downloading package stopwords to /root/nltk_data...
[nltk_data]   Package stopwords is already up-to-date!
[nltk_data] Downloading package wordnet to /root/nltk_data...
[nltk_data]   Package wordnet is already up-to-date!
[nltk_data] Downloading package punkt to /root/nltk_data...
[nltk_data]   Package punkt is already up-to-date!

Looking in indexes: https://pypi.org/simple, https://us-python.pkg.dev/colab-wheels/public/simple/
Collecting pyLDAvis
  Downloading pyLDAvis-3.4.0-py3-none-any.whl (2.6 MB)
     ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ 2.6/2.6 MB 25.8 MB/s eta 0:00:00
Requirement already satisfied: scipy in /usr/local/lib/python3.8/dist-packages (from pyLDAvis) (1.7.3)
Requirement already satisfied: scikit-learn>=1.0.0 in /usr/local/lib/python3.8/dist-packages (from pyLDAvis) (1.0.2)
Requirement already satisfied: setuptools in /usr/local/lib/python3.8/dist-packages (from pyLDAvis) (57.4.0)
Requirement already satisfied: jinja2 in /usr/local/lib/python3.8/dist-packages (from pyLDAvis) (3.1.2)
Requirement already satisfied: numpy>=1.22.0 in /usr/local/lib/python3.8/dist-packages (from pyLDAvis) (1.22.4)
Requirement already satisfied: pandas>=1.3.4 in /usr/local/lib/python3.8/dist-packages (from pyLDAvis) (1.3.5)
Requirement already satisfied: joblib>=1.2.0 in /usr/local/lib/python3.8/dist-packages (from pyLDAvis) (1.2.0)
Requirement already satisfied: gensim in /usr/local/lib/python3.8/dist-packages (from pyLDAvis) (3.6.0)
Collecting funcy
  Downloading funcy-2.0-py2.py3-none-any.whl (30 kB)
Requirement already satisfied: numexpr in /usr/local/lib/python3.8/dist-packages (from pyLDAvis) (2.8.4)
Requirement already satisfied: pytz>=2017.3 in /usr/local/lib/python3.8/dist-packages (from pandas>=1.3.4->pyLDAvis) (2022.7.1)
Requirement already satisfied: python-dateutil>=2.7.3 in /usr/local/lib/python3.8/dist-packages (from pandas>=1.3.4->pyLDAvis) (2.8.2)
Requirement already satisfied: threadpoolctl>=2.0.0 in /usr/local/lib/python3.8/dist-packages (from scikit-learn>=1.0.0->pyLDAvis) (3.1.0)
Requirement already satisfied: smart-open>=1.2.1 in /usr/local/lib/python3.8/dist-packages (from gensim->pyLDAvis) (6.3.0)
Requirement already satisfied: six>=1.5.0 in /usr/local/lib/python3.8/dist-packages (from gensim->pyLDAvis) (1.15.0)
Requirement already satisfied: MarkupSafe>=2.0 in /usr/local/lib/python3.8/dist-packages (from jinja2->pyLDAvis) (2.1.2)
Installing collected packages: funcy, pyLDAvis
Successfully installed funcy-2.0 pyLDAvis-3.4.0

Part One: Sentiment Modeling

# Reading the given data into dataframe
df = pd.read_csv("sentiment_analysis.csv") 
df.head(3)

	ID	text	label
0	7.680980e+17	Josh Jenkins is looking forward to TAB Breeder...	1
1	7.680980e+17	RT @MianUsmanJaved: Congratulations Pakistan o...	1
2	7.680980e+17	RT @PEPalerts: This September, @YESmag is taki...	1

print(df["label"].value_counts(normalize=True),'\n')
print(df.info())

1    0.674686
0    0.325314
Name: label, dtype: float64 

<class 'pandas.core.frame.DataFrame'>
RangeIndex: 550391 entries, 0 to 550390
Data columns (total 3 columns):
 #   Column  Non-Null Count   Dtype  
---  ------  --------------   -----  
 0   ID      550391 non-null  float64
 1   text    550391 non-null  object 
 2   label   550391 non-null  int64  
dtypes: float64(1), int64(1), object(1)
memory usage: 12.6+ MB
None

The data_frame contains 550391 rows and 3 columns, without non values. The text column contains the tweet content and The label column has the opinion : positive 1, negative 0. 67.46% positive opinion, 32% negative.

DF = df.copy(deep=True)
DF=DF.rename(columns={'text':'Text'})

# Defining a function which does all required text preprocessing on a given dataframe
def text_preprocessing(df):
  ''' Input DataFrame is required to have a column named "Text" 
      on which all pre-processing will be done'''

  DF=df.copy(deep=True)


  # Changing to lowercase 
  DF['Text'] = DF['Text'].apply(lambda x: " ".join(x.lower() for x in x.split()))

  # Removing special characters and words following @
  special_char=["rt","@\w*","http\S*","t.co\S*","[^a-zA-Z\s]","[^\w\s]"]
  for i in special_char:
    DF['Text'] = DF['Text'].str.replace(i,'')

  def language(sen):
    sen_t=sen.replace(' ','')
    if len(sen_t)>5:
      lang=detect_langs(sen)[0].lang
    else:
      lang='noten'
    return lang
  
  # Keeping examples only in English Language
  if ('Language' in (DF.columns)):
    idx=DF['Language']=='en'
    DF=DF[idx]  
  else:
    lang=DF['Text'].apply(lambda x:language(x))
    idx=lang=='en'
    DF=DF[idx]


  # Removing rows which have no non-whitespace characters   
  DF_T=DF.copy(deep=True)
  DF_T['Text'] = DF_T['Text'].str.replace(' ','')
  idx=(DF_T['Text'].apply(lambda x:len(x)))>0
  DF=DF[idx]

  # Removing stopwords
  stop_words = stopwords.words('english')
  DF['Text'] = DF['Text'].apply(lambda x: ' '.join(x for x in x.split() if x not in stop_words))


  # Removing rare words
  words= pd.Series(' '.join(DF['Text']).split()).value_counts()
  idx=words==1
  rare_words=words[idx]
  DF['Text'] = DF['Text'].apply(lambda x: ' '.join(x for x in x.split() if x not in rare_words))

  # Performing Tokenization
  DF['Text'] = DF['Text'].apply(lambda x: word_tokenize(x))

  stemmer = PorterStemmer()
  def stemming(L):
    for i in range(0,len(L)):
     L[i]=stemmer.stem(L[i])
    return L

  # Performing Stemming
  DF['Text'] = DF['Text'].apply(lambda x: stemming(x))

  return DF

'''  Below Commented Lines of codes are used to process raw data using 
above function - 'text_preprocessing'. 
Since, we have already run the processing, and saved the file, now we can just
upload the file and use it directly.
The processed file name is 'sentiment_analysis_pre-processed.csv' 
'''
#Processed_DF=text_preprocessing(DF)    
#Processed_DF=Processed_DF[['Text','label']]
#Processed_DF.to_csv('sentiment_analysis_pre-processed.csv', sep=',', index=False)
df = pd.read_csv("sentiment_analysis_pre-processed.csv") 
df

	Text	label
0	['josh', 'jenkin', 'look', 'forward', 'tab', '...	1
1	['congratul', 'pakistan', 'becom', 'notestteam...	1
2	['septemb', 'take', 'main', 'surpris', 'thanks...	1
3	['newli', 'paint', 'wall', 'thank', 'million',...	1
4	['excit', 'announc', 'juli', 'lab', 'reloc']	1
...	...	...
504691	['cant', 'stop', 'watch']	0
504692	['poor', 'old', 'tom', 'odel', 'doesnt', 'look...	0
504693	['antsmash', 'smash', 'ant', 'awesom']	1
504694	['morn', 'girl', 'wonder', 'friday']	1
504695	['bixbeat', 'mixtap', 'vol', 'great', 'aist', ...	1

504696 rows × 2 columns

Because of computational limitations, we will be dividing the above pre-processed dataframe into four subsets and train four classification models on these four subsets separately

# First, shuffling the above dataframe
df=pd.DataFrame(np.random.permutation(df),columns=df.columns)
y=df['label']
y=y.astype(float)
# Making four dataframes out of one pre-processed dataframe
df_1=df.iloc[0:126000,:]
y1=y[0:126000]

df_2=df.iloc[126000:252000,:]
y2=y[126000:252000]

df_3=df.iloc[252000:378000,:]
y3=y[252000:378000]

df_4=df.iloc[378000:504690,:]
y4=y[378000:504690]

# We will fit the TFIDF vectorizer on the complete dataframe
# But transformation will be done on the subset dataframes
# This will help maintain genralization of features across the four models
vectorizer = TfidfVectorizer(analyzer='word',token_pattern=r'\b[a-zA-Z]{3,}\b', min_df=100)
vectorizer.fit(df['Text'])

TfidfVectorizer(min_df=100, token_pattern='\\b[a-zA-Z]{3,}\\b')

1) Multinomial NAIVE BAYES

X=vectorizer.transform(df_1['Text']) # Using First Subset for Naive Bayes
X=X.toarray()
X = pd.DataFrame(data=X, columns=vectorizer.get_feature_names_out())
X_Train, X_Test, y_Train, y_Test = train_test_split(X,y1, test_size=0.20, random_state=10)

# Multinomial Naive Bayes
parameters = {'alpha':[0.01,0.05,0.1,0.5,1,5,10]}
model_NB= MultinomialNB()
clf = GridSearchCV(model_NB, parameters, cv=5, n_jobs=2)
#clf.fit(X_Train, y_Train)          ### No need to run again
best_score=92.26
best_param="'alpha':1"
print(f'The best cross-validation score: {round(best_score,2)}%')
print(f'The best parameter: '+best_param)

The best cross-validation score: 92.26%
The best parameter: 'alpha':1

bestmodel_NB=MultinomialNB(alpha=1.0)
bestmodel_NB.fit(X_Train, y_Train)
predictNB = bestmodel_NB.predict(X_Test)

test_accuracy = accuracy_score(y_Test, predictNB)
test_f1 = f1_score(y_Test, predictNB)

print(f'Test accuracy of Multinomial Naive Bayes model: {round(test_accuracy*100,2)}%')
print(f'Test f1 of Multinomial Naive Bayes model: {round(test_f1,2)}')

Test accuracy of Multinomial Naive Bayes model: 92.49%
Test f1 of Multinomial Naive Bayes model: 0.95

2) Logistic Regression

X=vectorizer.transform(df_2['Text']) # Using Second Subset for Logistic Regression
X=X.toarray()
X = pd.DataFrame(data=X, columns=vectorizer.get_feature_names_out())
X_Train, X_Test, y_Train, y_Test = train_test_split(X,y2, test_size=0.20, random_state=10)

parameters = {'C':[0.001,0.01, 0.1, 1, 5,20 ], 'penalty':['l1','l2']}

model_LR= LogisticRegression (solver='liblinear',max_iter=1000)
clf = GridSearchCV(model_LR, parameters, cv=5, n_jobs=2)
#clf.fit(X_Train, y_Train)     ### No need to run again
best_score=95.63
best_param="'C':5, 'penalty':'l2'"

print(f'The best cross-validation score: {round(best_score,2)}%')
print(f'The best parameter: '+best_param)

The best cross-validation score: 95.63%
The best parameter: 'C':5, 'penalty':'l2'

bestmodel_LR=LogisticRegression(max_iter=1000,C=5, solver= 'liblinear',penalty='l2')
bestmodel_LR.fit(X_Train, y_Train)
predictLR = bestmodel_LR.predict(X_Test)

test_accuracy = accuracy_score(y_Test, predictLR)
test_f1 = f1_score(y_Test, predictLR)

print(f'Test accuracy of Logestic Regression model: {round(test_accuracy*100,2)}%')
print(f'Test f1 score of Logestic Regression model: {round(test_f1,2)}')

Test accuracy of Logestic Regression model: 96.03%
Test f1 score of Logestic Regression model: 0.97

3) Linear Support Vector Machine

X=vectorizer.transform(df_3['Text']) # Using Third Subset for SVM
X=X.toarray()
X = pd.DataFrame(data=X, columns=vectorizer.get_feature_names_out())
X_Train, X_Test, y_Train, y_Test = train_test_split(X,y3, test_size=0.20, random_state=10)

parameters = {'C':[0.001, 0.01, 0.1, 1,5,10],'penalty':['l1','l2']}

model_SVC = LinearSVC(random_state=0, tol=1e-5,loss='squared_hinge',dual=False)
clf = GridSearchCV(model_SVC, parameters, cv=5, n_jobs=2)
#clf.fit(X_Train, y_Train)    ### No need to run again
best_score=95.66
best_param="'C':0.1, 'penalty':'l2'"

print(f'The best cross-validation score: {round(best_score,2)}%')
print(f'The best parameter: '+best_param)

The best cross-validation score: 95.66%
The best parameter: 'C':0.1, 'penalty':'l2'

bestmodel_SVC=LinearSVC(random_state=0, tol=1e-5,loss='squared_hinge',C = 0.1,penalty='l2',dual=False)
bestmodel_SVC.fit(X_Train, y_Train)
predictSVC = bestmodel_SVC.predict(X_Test)

test_accuracy = accuracy_score(y_Test, predictSVC)
test_f1 = f1_score(y_Test, predictSVC)

print(f'Test accuracy of SVC model: {round(test_accuracy*100,2)}%')
print(f'Test f1 score of SVC model: {round(test_f1,2)}')

Test accuracy of SVC model: 95.81%
Test f1 score of SVC model: 0.97

4) Random Forest

X=vectorizer.transform(df_4['Text']) # Using Fourth Subset for Random Forest
X=X.toarray()
X = pd.DataFrame(data=X, columns=vectorizer.get_feature_names_out())
X_Train, X_Test, y_Train, y_Test = train_test_split(X,y4, test_size=0.20, random_state=10)

parameters = {'n_estimators': [10,15,20],
              'min_samples_split': [100,200],
              'max_depth': [100,150]} 

model_RF = RandomForestClassifier(random_state=10)
clf = GridSearchCV(model_RF, parameters, cv=5, n_jobs=4)
#clf.fit(X_Train, y_Train)          ### No need to run again
best_score=92.94
best_param="'max_depth': 150, 'min_samples_split': 100, 'n_estimators': 20"

print(f'The best cross-validation score: {round(best_score,2)}%')
print(f'The best parameter: '+best_param)

The best cross-validation score: 92.94%
The best parameter: 'max_depth': 150, 'min_samples_split': 100, 'n_estimators': 20

bestmodel_RF = RandomForestClassifier(random_state=0, max_depth=150, n_estimators=20,min_samples_split=100)
bestmodel_RF.fit(X_Train, y_Train)
predictRF = bestmodel_RF.predict(X_Test)

test_accuracy = accuracy_score(y_Test, predictRF)
test_f1 = f1_score(y_Test, predictRF)

print(f'Test accuracy of RF model: {round(test_accuracy*100,2)}%')
print(f'Test f1 of RF model: {round(test_f1,2)}')

Test accuracy of RF model: 92.94%
Test f1 of RF model: 0.95

predictRF_train=bestmodel_RF.predict(X_Train)
accuracy_score(y_Train, predictRF_train)

0.9570704080827216

Next, we will prepare an ensemble from three of the above four models. Since, it is always better to have models in ensemble which are diverse in terms of algorithms. Therefore, we will have following three models in our ensemble,
1) Naive Bayes
2) Logistic Regression
3) Random Forest

# Preparing ensemble of three sentiment classifiers
CL_1 = MultinomialNB(alpha=1.0)
CL_2 = LogisticRegression(max_iter=1000,C=5, solver= 'liblinear',penalty='l2')
CL_4 = RandomForestClassifier(random_state=0, max_depth=150, n_estimators=20,min_samples_split=100)

est = [('Naive Bayes', CL_1), ('Logistic Regression', CL_2), ('Random Forest', CL_4)]
ens_sent_clf = VotingClassifier(estimators=est, voting='hard',n_jobs=1)

Now just for reference, We will train our ensemble model using half of the entire pre-processed dataframe instead of the subsets, and check the test set accuracy

df=pd.DataFrame(np.random.permutation(df),columns=df.columns)
h=int(df.shape[0]*0.5)
df_h=df.iloc[0:h,:]
y=df_h['label']
y=y.astype(float)

# Making Training Set and Test Set for use with ensemble model
X=vectorizer.transform(df_h['Text'])
X=X.toarray()
X = pd.DataFrame(data=X, columns=vectorizer.get_feature_names_out())

# Making Training and Test set manually because of RAM limitation
t=int((X.shape[0])*0.8)
X_Train=X.iloc[0:t,:]
y_Train=y[0:t]
X_Test=X.iloc[t:,:]
y_Test=y[t:]

''' Not Running the below fit again, instead hard coding the results'''
#ens_sent_clf.fit(X_Train, y_Train)
#predict_ens = ens_sent_clf.predict(X_Test)
#test_accuracy = accuracy_score(y_Test, predict_ens)
#test_f1 = f1_score(y_Test, predict_ens)
test_accuracy=0.9569
test_f1=0.97

print(f'Test accuracy of Ensemble Model: {round(test_accuracy*100,2)}%')
print(f'Test f1 of Ensemble Model: {round(test_f1,2)}')

Test accuracy of Ensemble Model: 95.69%
Test f1 of Ensemble Model: 0.97

It is to be noted that the dataset used for training the above ensemble model does not have examples related to the context of the application, which is Russia-Ukraine Conflict.
It is therefore of utmost importance to include enough examples in training data related to the application of the model.

Our strategy for model training is proposed as below:

1) Only include 80K examples from the given dataset for part 1 of project : 'sentiment_analysis', with 40K examples of positive sentiment tweets, and 40K examples of negative sentiment tweets, to give a general idea of positive and negative sentiment examples to the model.

2) We will generate another training dataset from twitter datasets related to Russia-Ukraine War. Since, these datasets will not be labelled in their raw form, we will label them using below strategy:
(i) Label the tweets containing the words 'kill,crime,war, violence, invasion, attack,conflict,evil,weapon,missile,lies, fight,shelling,genocide,death' to be negative sentiment tweets.It is interesting to note that, major proportion of these negative sentiment tweets can be regarded as Pro-Ukrainian Tweets. For this purpose, we will use #StandwithUkraine twitter dataset from Kaggle.

(ii) To build a well trained classifier with respect to the task at hand, we need examles of positive sentiment tweets. For this, we shall combine examples with #standwithrussia #istandwithrussia #standwithputin #istandwithputin #stopnato For this purpose, we use 'The Russian invasion of Ukraine without retweets' dataset from Kaggle. We will filter the tweets with above hashtags from this dataset

Preparing Training Data

# Part 1 from proposal above
# From given dataset for Part 1, 'sentiment_analysis.csv'
# Taking 40K examples for positive sentiment , i.e. with label 1
# Taking 40K examples for negative sentiment, i.e. with label 0
df = pd.read_csv("sentiment_analysis.csv") 
df=pd.DataFrame(np.random.permutation(df),columns=df.columns)
idx=df['label']==1
df_1=(df[idx]).iloc[0:40000]
idx=df['label']==0
df_2=(df[idx]).iloc[0:40000]
df_tr1=pd.concat([df_1,df_2])
df_tr1=pd.DataFrame(np.random.permutation(df_tr1),columns=df.columns)
df_tr1=df_tr1.rename(columns={'text': 'Text'})
df_tr1=df_tr1.reset_index()
df_tr1=df_tr1[['Text','label']]

# Defining function which removs the words that are expected to be in majority of the hashtags
# because this might mislead the model
def remove_gen_keywords(df):
  df['Text'] = df['Text'].str.replace('StandWithUkraine','')  
  df['Text'] = df['Text'].str.replace('standwithukraine','')  
  df['Text'] = df['Text'].str.replace('standwithUkraine','')  
  df['Text'] = df['Text'].str.replace('Ukraine','')  
  df['Text'] = df['Text'].str.replace('ukraine','')  
  df['Text'] = df['Text'].str.replace('russia','')  
  df['Text'] = df['Text'].str.replace('Ukrainian','') 
  df['Text'] = df['Text'].str.replace('ukrainian','')
  df['Text'] = df['Text'].str.replace('Russian','') 
  df['Text'] = df['Text'].str.replace('stand','') 
  df['Text'] = df['Text'].str.replace('Stand','') 
  return df

# Part 2 (i) from proposal above
# This training data represent negative sentiment in context of 
# Russia-Ukraine War
keywords=['kill','crim','conflict', 'violence','evil','innocent','die',
          'missile','shell','genocide','death','lives','lie','suffer'
          'invade','dict','stop','enough','end','invasion','fear','pain'
          'resis','offens','soldier','sacrif','atroci','occup','brutal'
          'help','children','food','please','cri','fight','attack','ang'
          'blood','murder','victim','bomb','terror','aggre','human']
df_tr2=pd.read_csv('Twitter_Kaggle_StandWithUkraine.csv')
df_text=df_tr2['content']
idx=df_text.str.contains(keywords[0])
for i in range(1,len(keywords)):
  idx=df_text.str.contains(keywords[i])+idx
df_tr2=df_tr2[idx]
df_tr2['label']=np.zeros(len(df_tr2))
df_tr2=df_tr2.rename(columns={'content': 'Text'})
df_tr2=df_tr2.reset_index()
df_tr2=df_tr2[['Text','label']]
df_tr2=remove_gen_keywords(df_tr2)

# Part 2 (ii) from proposal above
# This training data represent positive sentiment in context of 
# Russia-Ukraine War
# Or in other words Pro-Russian Tweets
df_tr3=pd.read_csv('The Russian invasion of Ukraine without retweets.csv')
df_tr3['Text'] = df_tr3['text'].apply(lambda x: " ".join(x.lower() for x in x.split()))
df_text=df_tr3['Text']
idx_1=df_text.str.contains('standwithrussia')
idx_2=df_text.str.contains('standwithputin')
idx_3=df_text.str.contains('naziukraine')
idx_4=df_text.str.contains('stopnato')
idx=idx_1|idx_2|idx_3|idx_4
df_tr3=df_tr3[idx]
df_tr3['label']=np.ones(len(df_tr3))
df_tr3=df_tr3.reset_index()
df_tr3=df_tr3[['Text','label']]

# Final Training Dataset
df_training=pd.concat([df_tr1,df_tr2,df_tr3])
df_training=pd.DataFrame(np.random.permutation(df_training),columns=df_training.columns)
df_training=df_training.reset_index()
df_training['label'].astype(float)
df_training=df_training[['Text','label']]

#Processed_trainDF=text_preprocessing(df_training)     ### No need to run pre-processing again
#Processed_trainDF=Processed_trainDF[['Text','label']]
#Processed_trainDF.to_csv('Processed_trainDF.csv', sep=',', index=False)
df_training = pd.read_csv("Processed_trainDF.csv")   ### Reading processed csv file

Note that we will not be splitting the above training data into training set and testing set. This is because we will test the trained model on another UNSEEN dataset from twitter called '30K Tweets with russiaukrainewar hashtag' and compare the predictions on this dataset from our model and a pre-trained NLP model from Google Cloud Platform.

# Preparing TF-IDF Features from Prepared Training Dataset

vectorizer = TfidfVectorizer(analyzer='word',token_pattern=r'\b[a-zA-Z]{3,}\b', min_df=50)
vectorizer.fit(df_training['Text'])

X_Train=vectorizer.transform(df_training['Text'])
X_Train=X_Train.toarray()
X_Train = pd.DataFrame(data=X_Train, columns=vectorizer.get_feature_names_out())
y_Train=df_training['label']

# Training our best ensemble model
ens_sent_clf.fit(X_Train, y_Train)

# Calculating accuracy on training set just for reference
predict_ens = ens_sent_clf.predict(X_Train)
train_accuracy = accuracy_score(y_Train, predict_ens)
train_f1 = f1_score(y_Train, predict_ens)

print(f'Training accuracy of Our Ensemble Model: {round(train_accuracy*100,2)}%')
print(f'Trainig f1 score of Our Ensemble Model: {round(train_f1,2)}')

Training accuracy of Our Ensemble Model: 95.61%
Trainig f1 score of Our Ensemble Model: 0.94

Part Two: Sentiment classification

Now that we have trained our best model with relevant training data,
Next task is to make prediction on an UNSEEN twitter data '30K Tweets with russiaukrainewar hashtag', and also comparing these predictions with a pre-trained NLP model from Google Cloud Platform

# Reading and Pre-Processing Unseen Test Data

df_test = pd.read_csv("30K Tweets with russiaukrainewar hashtag.csv")    
df_test=df_test.rename(columns={'Tweet': 'Text'})          
Processed_testDF_1=text_preprocessing(df_test)            

Processed_testDF_1=Processed_testDF_1[['Text','Location','Time']]           

X_Test=vectorizer.transform(Processed_testDF_1['Text'].astype(str))
X_Test=X_Test.toarray()
X_Test = pd.DataFrame(data=X_Test, columns=vectorizer.get_feature_names_out())

# Making Pediction using our trained model on this test dataset
predict_ens = ens_sent_clf.predict(X_Test)

Now testing the same test dataset using Google Cloud Natural Language API

from google.cloud import language_v1
from google.oauth2.service_account import Credentials

credentials = Credentials.from_service_account_file('/content/nlp-sentiment-analysis-project-f1178a489013.json')
'''Note that if this notebook is not being run on Google Colab,
  then the above path to the json key file needs to be changed accordingly '''
client = language_v1.LanguageServiceClient(credentials=credentials)

Strategy to analyse tweets using the Google Natural Language API

Since, we need to analyse thousands of tweets using the Natural Language API, we cannot make that many API calls. We will utilize the API's feature that it gives a separate sentiment score for each sentence in the document.
Therefore, we will modify our tweets dataset such that:
1) Multiple sentences in a single tweet become one sentence, i.e. we remove the period symbol '.' within the tweets
2) Join any two tweets by '.' so that the API considers each tweet as one sentence.
Thus, we will pass a batch of tweets as one document in a single API call, and then get sentiment scores for each tweet as sentiment scores for sentences.

# Defining a function which does the pre-processing as explained above
# to feed tweets data to the API
def preprocessing_for_NLP_API(df):
  ''' Input DataFrame is required to have a column named "Text" 
      on which all pre-processing will be done'''

  DF=df.copy(deep=True)

  # Changing to lowercase 
  DF['Text'] = DF['Text'].apply(lambda x: " ".join(x.lower() for x in x.split()))

  # Removing unwanted patterns in the tweets
  special_char=["rt","@\w*","http\S*","t.co\S*","[^a-zA-Z\s]","[^\w\s]"]
  for i in special_char:
    DF['Text'] = DF['Text'].str.replace(i,'')

  def language(sen):
    sen_t=sen.replace(' ','')
    if len(sen_t)>5:
      lang=detect_langs(sen)[0].lang
    else:
      lang='noten'
    return lang
  
  # Keeping examples only in English Language
  if ('Language' in (DF.columns)):
    idx=DF['Language']=='en'
    DF=DF[idx]  
  else:
    lang=DF['Text'].apply(lambda x:language(x))
    idx=lang=='en'
    DF=DF[idx]  
    DF=DF.reset_index()

  # Removing rows which have no non-whitespace characters   
  DF_T=DF.copy(deep=True)
  DF_T['Text'] = DF_T['Text'].str.replace(' ','')
  idx=(DF_T['Text'].apply(lambda x:len(x)))>0
  DF=DF[idx]
  DF=DF.reset_index()

  # Joining all tweets
  l=DF['Text']
  document=' the. The '.join(l)

  return document

# Defining Function which takes in a test dataframe of tweets
# and return predictions made by Natural Language API

def Predict_Natural_Language_API(df_test,batch_size):
  l=df_test.shape[0]
  idx=np.arange(0,l,batch_size)
  idx=np.append(idx,l)
  scores=[]
  
  for i in range(0,len(idx)-1):
    document_=preprocessing_for_NLP_API(df_test.iloc[idx[i]:idx[i+1],:])
    document = language_v1.Document(content=document_, type_=language_v1.Document.Type.PLAIN_TEXT)
    response = client.analyze_sentiment(request={'document': document})
    new_scores=[response.sentences[i].sentiment.score for i in range(0,len(response.sentences))]
    scores=np.append(scores,new_scores)
    # Considering tweets with sentiment score > 0 as positive sentiment tweets,
    sentiment_predict=(scores>0).astype(int)
  return sentiment_predict,scores

predictions_API,scores=Predict_Natural_Language_API(df_test,3000)

count_api=collections.Counter(predictions_API)
count_ens_model=collections.Counter(predict_ens)

plt.figure(figsize=[6,5])
plt.pie([count_api[0],count_api[1]],labels=['Negative Sentiment','Positive Sentiment'])
cir=plt.Circle( (0,0), 0.7, color='white')
p=plt.gcf()
p.gca().add_artist(cir)
plt.title('Sentiment Distribution of Predictions by NL API\n \nDataset:30K Tweets with #russiaukrainewar , Kaggle')
plt.show()

plt.figure(figsize=[6,5])
plt.pie([count_ens_model[0],count_ens_model[1]],labels=['Negative Sentiment','Positive Sentiment'])
cir=plt.Circle( (0,0), 0.7, color='white')
p=plt.gcf()
p.gca().add_artist(cir)
plt.title('Sentiment Distribution of' 
' Predictions by Our Trained Model\n'
' \n Dataset:30K Tweets with #russiaukrainewar , Kaggle \n'
'(Not Used for Training)')
plt.show()

Now, assuming that the predictions made by Natural Language API are true values or true labels, we shall calculate accuracy of our ensemble model as below,

Benchmarking with Natural Language API,

idx=(scores<-0.3 ) | (scores>0.3 ) 
# Ignoring the Neutral Tweets for comparison of Accuracy
# As our model predicts only positive or negative sentiments  
predict_api_wo_neutral=predictions_API[idx]
predict_model_wo_neutral=predict_ens[idx]
accuracy_score_benchmark=accuracy_score(predict_api_wo_neutral, predict_model_wo_neutral)
print(f'Accuracy with respect to Natural Language API is : {round(accuracy_score_benchmark*100,2)}%')

Accuracy with respect to Natural Language API is : 82.88%

Thus, we get about 83% match with Natural Language API on the '30K Tweets with russiaukrainewar hashtag' dataset from Kaggle which is not used anywhere in the model training.

Observation and Discussion:

We can see that the majority of sentiments in the gathered data are negative. This suggests that people have a generally unfavorable view of the situation in Ukraine. In the context of the Russian invasion of Ukraine, this could mean that the actions of Russia and the impact of the conflict on the Ukrainian people have caused negative emotions and opinions towards the situation.

On the other hand, the small percentage of positive sentiments could suggest that there are some positive developments or events related to the conflict that have been reported. For example, positive sentiments could be related to instances of aid or support for Ukraine or any developments that could lead to a peaceful resolution of the conflict.

Overall, the fact that positive sentiments only account for 25% of the total sentiments suggests that not only people have a strongly negative perception of the situation in Ukraine but also Ukraine is not getting the proper attention, support, and aid needed.

Please note that, until this point we have only trained and validated our model, analysis work is yet to be done using our trained model on Relatively Large Analysis Test Dataset from time period across the year 2022.

Test data predictions

"""No need to execute the web scraping code, as the data has already been obtained and uploaded from a CSV file"""
"""Twitter data web scraping code"""
# !pip install snscrape
# import pandas as pd
# import snscrape.modules.twitter as sntwitter
# import datetime
# import timeit
# query = '#Ukrainewar OR #StandWithUkraine OR #StandWithRussia OR #standwithputin OR #UkraineRussiaWar OR #Nato OR #standwithzelensky OR #Ukraine OR #RussianAggression OR #UnitedForUkraine OR #Crimea OR #Donbass OR #WarInDonbass OR #IStandWithPutin OR #RussiaUkraineConflict'
# start_date = datetime.date(2022, 2, 20)
# end_date = datetime.date(2023, 3, 31)
# end_date-start_date

# start = timeit.default_timer()
# count = 0
# tweets = []
# delta = datetime.timedelta(days=1)
# for day in range((end_date - start_date).days + 1):
#     count+=1
#     print("Day: ", count)
#     tweets_day = []
#     since_date = (start_date + datetime.timedelta(days=day)).strftime('%Y-%m-%d')
#     until_date = (start_date + datetime.timedelta(days=day+1)).strftime('%Y-%m-%d')
#     query_day = f'{query} since:{since_date} until:{until_date}'
#     scraper = sntwitter.TwitterSearchScraper(query_day)
#     for i, tweet in enumerate(scraper.get_items()):
#         if i >= 1000:
#             break
#         data = [tweet.date, tweet.user.username, tweet.id, tweet.content, tweet.lang, tweet.user.location, tweet.coordinates, tweet.likeCount, tweet.retweetCount, tweet.hashtags, tweet.url]
#         tweets_day.append(data)
#     tweets.append(pd.DataFrame(tweets_day, columns=['Date', 'Username', 'Id', 'Text', 'Language', 'Location', 'Coordinates', 'Likes', 'Retweets', 'Hashtags', 'URL']))
#     clear_output(wait=True)


# df = pd.concat(tweets).reset_index(drop=True)

# stop = timeit.default_timer()
# print('Time: ', stop - start)

# df.to_csv('Twitter data ' + str(start_date) + ' to ' + str(end_date) + '.csv')

'''Reading the twitter data'''
# twitter_data = pd.read_csv("Twitter data.csv") ### No need to run as we are directly uploading the text pre-processed csv file
# twitter_data

'''Function that performs the necessary text processing on a given dataframe'''
def text_preprocessing_tweet(df):
   ''' Input DataFrame is required to have a column named "Text" 
       on which all pre-processing will be done'''

   DF=df.copy(deep=True)
   DF['Tweet'] = df.loc[:, 'Text']


   # Changing to lowercase 
   DF['Text'] = DF['Text'].apply(lambda x: " ".join(x.lower() for x in x.split()))

   # Removing special characters and words following @
   special_char=["rt","@\w*","http\S*","t.co\S*","[^a-zA-Z\s]","[^\w\s]"]
   for i in special_char:
     DF['Text'] = DF['Text'].str.replace(i,'')

   def language(sen):
     sen_t=sen.replace(' ','')
     if len(sen_t)>5:
       lang=detect_langs(sen)[0].lang
     else:
       lang='noten'
     return lang
  
   # Keeping examples only in English Language
   if ('Language' in (DF.columns)):
     idx=DF['Language']=='en'
     DF=DF[idx]  
   else:
     lang=DF['Text'].apply(lambda x:language(x))
     idx=lang=='en'
     DF=DF[idx]


   # Removing rows which have no non-whitespace characters   
   DF_T=DF.copy(deep=True)
   DF_T['Text'] = DF_T['Text'].str.replace(' ','')
   idx=(DF_T['Text'].apply(lambda x:len(x)))>0
   DF=DF[idx]

   # Removing rows which have no non-whitespace characters in the tweet column   
   DF_T=DF.copy(deep=True)
   DF_T['Tweet'] = DF_T['Tweet'].str.replace(' ','')
   idx=(DF_T['Tweet'].apply(lambda x:len(x)))>0
   DF=DF[idx]

   # Removing stopwords
   stop_words = stopwords.words('english')
   DF['Text'] = DF['Text'].apply(lambda x: ' '.join(x for x in x.split() if x not in stop_words))


   # Removing rare words
   words= pd.Series(' '.join(DF['Text']).split()).value_counts()
   idx=words==1
   rare_words=words[idx]
   DF['Text'] = DF['Text'].apply(lambda x: ' '.join(x for x in x.split() if x not in rare_words))

   # Performing Tokenization
   DF['Text'] = DF['Text'].apply(lambda x: word_tokenize(x))

   stemmer = PorterStemmer()
   def stemming(L):
     for i in range(0,len(L)):
      L[i]=stemmer.stem(L[i])
     return L

   # Performing Stemming
   DF['Text'] = DF['Text'].apply(lambda x: stemming(x))

   # Removing empty list
   DF = DF[DF['Text'] != '[]']

   return DF

"""No need to execute the text preprocessing code, as the text preprocessing has already been performed and uploaded from a CSV file"""
# Pre-processing the twitter data
# twitter_data = twitter_data.drop(['Index', 'Id'], axis = 1)
# df = text_preprocessing_tweet(twitter_data)
# df.to_csv('twitter_data_pre-processed.csv', sep=',', index=False)
# df

'No need to execute the text preprocessing code, as the text preprocessing has already been performed and uploaded from a CSV file'

# Reading the text preprocessed twitter data file
df = pd.read_csv('twitter_data_pre-processed.csv')

# Removes any rows where the 'Text' column contains an empty list
df = df[df['Text'] != '[]']
df

	Date	Username	Text	Language	Location	Coordinates	Likes	Retweets	Hashtags	URL	Tweet
0	2022-02-20 23:59:59+00:00	d801227313a94cc	['bbc', 'news', 'ukrain', 'russia', 'possibl',...	en	NaN	NaN	1	0	['Ukraine', 'Russia']	https://twitter.com/d801227313a94cc/status/149...	BBC News - #Ukraine: What are #Russia's possib...
1	2022-02-20 23:59:58+00:00	ttnewmm	['nato', 'usa', 'uk', 'ukrain', 'usa', 'uk', '...	en	NaN	NaN	2	1	['NATO', 'USA', 'UK', 'Ukraine']	https://twitter.com/ttnewmm/status/14955488150...	#NATO #USA #UK #Ukraine THE USA - UK - NATO NE...
2	2022-02-20 23:59:39+00:00	ban239	['happi', 'paicip', 'mani', 'other', 'frigid',...	en	Washington, DC, Kraków	Coordinates(longitude=-77.144435, latitude=38....	4	0	['StandWithUkraine']	https://twitter.com/ban239/status/149554873720...	Happy to have participated with so many others...
3	2022-02-20 23:59:29+00:00	JustAMazen	['perhap', 'us', 'invad', 'canada', 'annex', '...	en	NaN	NaN	8	2	['StandWithUkraine', 'PrayForUkraine']	https://twitter.com/JustAMazen/status/14955486...	@OlgaNYC1211 Perhaps when US invades Canada, a...
4	2022-02-20 23:59:17+00:00	em_az	['yep', 'spanish', 'flu', 'call', 'told', 'put...	en	US	NaN	0	1	['PutinsWar', 'UkraineRussiaCrisis', 'UkraineI...	https://twitter.com/em_az/status/1495548645889...	@mikecalma Yep. The Spanish flu called. Told P...
...	...	...	...	...	...	...	...	...	...	...	...
249123	2023-03-29 18:21:41+00:00	job360ng	['current', 'hire', 'sma', 'passion', 'individ...	en	Lagos, Nigeria	NaN	1	0	['HIRINGNOW', 'lagosjobs', 'certificateofretur...	https://twitter.com/job360ng/status/1641143608...	We are currently hiring smart and passionate i...
249124	2023-03-29 18:21:39+00:00	lewski5800	['trump', 'vladimir', 'putin', 'trump', 'putin...	en	NaN	NaN	0	0	['trump', 'putin', 'ukraine', 'ninetyeight', '...	https://twitter.com/lewski5800/status/16411435...	Trump On Vladimir Putin 🤯 #trump #putin #ukrai...
249125	2023-03-29 18:21:30+00:00	f_o_r_Ukraine	['happi', 'morn', 'russian', 'occupi', 'melito...	en	Ukraine	NaN	4	1	['UkraineWar']	https://twitter.com/f_o_r_Ukraine/status/16411...	Not very happy morning for Russians in occupie...
249126	2023-03-29 18:21:24+00:00	IvanIhorovych	['break', 'ukrain', 'stori', 'deni', 'heard', ...	en	NaN	NaN	0	0	['BREAKING', 'Ukraine', 'CrimesAgainstHumanity...	https://twitter.com/IvanIhorovych/status/16411...	#BREAKING #Ukraine\nThe story of Denys, who he...
249127	2023-03-29 18:21:19+00:00	GordonGChang	['talk', 'london', 'time', 'china', 'diplomaci...	en	NaN	NaN	22	4	['China', 'Taiwan', 'Russia', 'Ukraine', 'Hond...	https://twitter.com/GordonGChang/status/164114...	Talking with @Nigel_Farage at 7:30 London time...

249115 rows × 11 columns

# Checking null values
print(df.shape)
df['Text'].isna().value_counts()

(249115, 11)

False    249115
Name: Text, dtype: int64

# Transforming the 'Text' column
X_test_tweet = vectorizer.transform(df['Text'].astype(str))
X_test_tweet = X_test_tweet.toarray()
X_test_tweet = pd.DataFrame(data=X_test_tweet, columns=vectorizer.get_feature_names_out())

# Predicting the test data(twitter)
predict_ens_tweet = ens_sent_clf.predict(X_test_tweet)
predict_ens_tweet
# print(len(predict_ens_tweet))

array([0, 0, 0, ..., 1, 0, 1])

# Count of each prediction
count_ens_model_tweet=collections.Counter(predict_ens_tweet)
count_ens_model_tweet

Counter({0: 195180, 1: 53935})

# Plotting the pie chart to visualize the sentiment analysis results of the test data 
plt.figure(figsize=[6,5])
total = total_1 = count_ens_model_tweet[0] + count_ens_model_tweet[1]
plt.pie([count_ens_model_tweet[0],count_ens_model_tweet[1]],labels=['{}%'.format(round(count_ens_model_tweet[0]/total_1, 2)*100),'{}%'.format(round(count_ens_model_tweet[1]/total, 2)*100)])
cir=plt.Circle( (0,0), 0.7, color='white')
p=plt.gcf()
p.gca().add_artist(cir)
plt.title('Sentiment Distribution of Predictions Ensemble model for tweets representing entire war period')
plt.legend(['Negative Sentiment', 'Positive Sentiment'], loc='upper right', bbox_to_anchor=(2, 1))
plt.show()

# Function that performs the necessary text processing for generating a wordcloud from tweets
def process_wordcloud_tweet(df):
  ''' Input DataFrame is required to have a column named "Tweet" 
      on which all pre-processing will be done'''

  DF=df.copy(deep=True)


  # Changing to lowercase 
  DF['Tweet_processed'] = DF['Tweet'].apply(lambda x: " ".join(x.lower() for x in x.split()))

  # Removing special characters and words following @
  special_char=["rt","@\w*","http\S*","t.co\S*","[^a-zA-Z\s]","[^\w\s]"]
  for i in special_char:
    DF['Tweet_processed'] = DF['Tweet_processed'].str.replace(i,'')

  def language(sen):
    sen_t=sen.replace(' ','')
    if len(sen_t)>5:
      lang=detect_langs(sen)[0].lang
    else:
      lang='noten'
    return lang

  # Keeping examples only in English Language
  if ('Language' in (DF.columns)):
    idx=DF['Language']=='en'
    DF=DF[idx]  
  else:
    lang=DF['Tweet_processed'].apply(lambda x:language(x))
    idx=lang=='en'
    DF=DF[idx]

  DF=remove_gen_keywords(DF)

  # Removing rows which have no non-whitespace characters   
  DF_T=DF.copy(deep=True)
  DF_T['Tweet_processed'] = DF_T['Tweet_processed'].str.replace(' ','')
  idx=(DF_T['Tweet_processed'].apply(lambda x:len(x)))>0
  DF=DF[idx]

  # Removing stopwords
  stop_words = stopwords.words('english')
  DF['Tweet_processed'] = DF['Tweet_processed'].apply(lambda x: ' '.join(x for x in x.split() if x not in stop_words))

  # Removing rare words
  words= pd.Series(' '.join(DF['Tweet_processed']).split()).value_counts()
  idx=words==1
  rare_words=words[idx]
  DF['Tweet_processed'] = DF['Tweet_processed'].apply(lambda x: ' '.join(x for x in x.split() if x not in rare_words))
 
  return DF

# Separating positive and negative tweets from the above dataset
idx_ukr=predict_ens_tweet==0
idx_rus=predict_ens_tweet==1
df_word_cloud_tweet=process_wordcloud_tweet(df)
df_ukr_tweet=(df_word_cloud_tweet[idx_ukr]).astype(str)
df_rus_tweet=(df_word_cloud_tweet[idx_rus]).astype(str)
df_neg = df[idx_ukr]
df_pos = df[idx_rus]

# Renaming and splitting the 'Date' column for performing further analysis in daraframe df_word_cloud_tweet
df_word_cloud_tweet.rename(columns={'Date': 'Datetime'}, inplace=True)
df_word_cloud_tweet['Date'] = df_word_cloud_tweet['Datetime'].str.split(' ')
df_word_cloud_tweet["Date"] = df_word_cloud_tweet["Date"].str[0]

# Preprocessing positive and negative tweet data and generating seperate word cloud for each category
a=df_ukr_tweet['Tweet_processed']
text_ukr_tweet=' '.join(a)
b=df_rus_tweet['Tweet_processed']
text_rus_tweet=' '.join(b)
word_cloud_ukr_tweet = WordCloud(collocations = False, background_color = 'white').generate(text_ukr_tweet)
word_cloud_rus_tweet = WordCloud(collocations = False, background_color = 'white').generate(text_rus_tweet)

#Plotting WordCloud from Pro-Ukrainian Tweets (or Tweets with Negative Sentiment)
plt.figure(figsize=[12,10])
plt.imshow(word_cloud_ukr_tweet, interpolation='bilinear')
plt.axis("off")
plt.show()

#Plotting WordCloud from Pro-Russian Tweets (or Tweets with Positive Sentiment)
plt.figure(figsize=[12,10])
plt.imshow(word_cloud_rus_tweet, interpolation='bilinear')
plt.axis("off")
plt.show()

df

	Datetime	Username	Text	Language	Location	Coordinates	Likes	Retweets	Hashtags	URL	Tweet	Datetime	prediction	positive_sentiment	negative_sentiment
0	2022-02-20 23:59:59+00:00	d801227313a94cc	['bbc', 'news', 'ukrain', 'russia', 'possibl',...	en	NaN	NaN	1	0	['Ukraine', 'Russia']	https://twitter.com/d801227313a94cc/status/149...	BBC News - #Ukraine: What are #Russia's possib...	2022-02-20	0	0	1
1	2022-02-20 23:59:58+00:00	ttnewmm	['nato', 'usa', 'uk', 'ukrain', 'usa', 'uk', '...	en	NaN	NaN	2	1	['NATO', 'USA', 'UK', 'Ukraine']	https://twitter.com/ttnewmm/status/14955488150...	#NATO #USA #UK #Ukraine THE USA - UK - NATO NE...	2022-02-20	0	0	1
2	2022-02-20 23:59:39+00:00	ban239	['happi', 'paicip', 'mani', 'other', 'frigid',...	en	Washington, DC, Kraków	Coordinates(longitude=-77.144435, latitude=38....	4	0	['StandWithUkraine']	https://twitter.com/ban239/status/149554873720...	Happy to have participated with so many others...	2022-02-20	0	0	1
3	2022-02-20 23:59:29+00:00	JustAMazen	['perhap', 'us', 'invad', 'canada', 'annex', '...	en	NaN	NaN	8	2	['StandWithUkraine', 'PrayForUkraine']	https://twitter.com/JustAMazen/status/14955486...	@OlgaNYC1211 Perhaps when US invades Canada, a...	2022-02-20	0	0	1
4	2022-02-20 23:59:17+00:00	em_az	['yep', 'spanish', 'flu', 'call', 'told', 'put...	en	US	NaN	0	1	['PutinsWar', 'UkraineRussiaCrisis', 'UkraineI...	https://twitter.com/em_az/status/1495548645889...	@mikecalma Yep. The Spanish flu called. Told P...	2022-02-20	0	0	1
...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...
249110	2023-03-29 18:21:41+00:00	job360ng	['current', 'hire', 'sma', 'passion', 'individ...	en	Lagos, Nigeria	NaN	1	0	['HIRINGNOW', 'lagosjobs', 'certificateofretur...	https://twitter.com/job360ng/status/1641143608...	We are currently hiring smart and passionate i...	2023-03-29	1	1	0
249111	2023-03-29 18:21:39+00:00	lewski5800	['trump', 'vladimir', 'putin', 'trump', 'putin...	en	NaN	NaN	0	0	['trump', 'putin', 'ukraine', 'ninetyeight', '...	https://twitter.com/lewski5800/status/16411435...	Trump On Vladimir Putin 🤯 #trump #putin #ukrai...	2023-03-29	0	0	1
249112	2023-03-29 18:21:30+00:00	f_o_r_Ukraine	['happi', 'morn', 'russian', 'occupi', 'melito...	en	Ukraine	NaN	4	1	['UkraineWar']	https://twitter.com/f_o_r_Ukraine/status/16411...	Not very happy morning for Russians in occupie...	2023-03-29	1	1	0
249113	2023-03-29 18:21:24+00:00	IvanIhorovych	['break', 'ukrain', 'stori', 'deni', 'heard', ...	en	NaN	NaN	0	0	['BREAKING', 'Ukraine', 'CrimesAgainstHumanity...	https://twitter.com/IvanIhorovych/status/16411...	#BREAKING #Ukraine\nThe story of Denys, who he...	2023-03-29	0	0	1
249114	2023-03-29 18:21:19+00:00	GordonGChang	['talk', 'london', 'time', 'china', 'diplomaci...	en	NaN	NaN	22	4	['China', 'Taiwan', 'Russia', 'Ukraine', 'Hond...	https://twitter.com/GordonGChang/status/164114...	Talking with @Nigel_Farage at 7:30 London time...	2023-03-29	1	1	0

249115 rows × 15 columns

# Renaming and splitting the 'Date' column for performing further analysis in dataframe df
df.rename(columns={'Date': 'Datetime'}, inplace=True)
df['Date'] = df['Datetime'].str.split(' ')
df["Date"] = df["Date"].str[0]

# Adding the 'prediction','positive_sentiment' and 'negative_sentiment' column to the dataframe df
df['prediction']= predict_ens_tweet
df['positive_sentiment'] = predict_ens_tweet
temp = predict_ens_tweet+1
df['negative_sentiment'] = temp
df.loc[(df.negative_sentiment == 2),'negative_sentiment']=0

# Getting the count of samples in each day
start_date = datetime.date(2022, 2, 20)
end_date = datetime.date(2023, 3, 29)
N = []
for delta in range((end_date - start_date).days + 1):
    day = start_date + datetime.timedelta(delta)
    N.append(df['Date'].value_counts()[str(day)])
# print(N)

# Sentiment analysis for each day
positive_sentiment = []
negative_sentiment = []
a = 0
n = 0
temp = []
for i in N:
    n = n+i
    temp.append(df[a:n].shape[0])
    day_predictions = df['prediction'][a:n]
    count_predictions = day_predictions.value_counts()
    negative_sentiment.append(count_predictions.get(0, 0))
    positive_sentiment.append(count_predictions.get(1, 0))
    a = n

[509, 465, 554, 503, 494, 538, 510, 482, 513, 540, 534, 568, 565, 611, 563, 619, 571, 539, 551, 549, 538, 519, 530, 582, 563, 577, 563, 518, 526, 534, 518, 508, 556, 490, 518, 520, 541, 512, 581, 502, 536, 533, 564, 549, 517, 549, 551, 532, 548, 547, 534, 570, 492, 473, 511, 488, 492, 531, 565, 538, 544, 522, 543, 503, 559, 534, 532, 511, 527, 464, 515, 501, 518, 508, 444, 490, 497, 501, 519, 553, 525, 537, 537, 288, 537, 499, 493, 506, 503, 483, 508, 487, 541, 518, 575, 524, 499, 484, 453, 523, 464, 514, 476, 502, 527, 461, 516, 438, 506, 478, 497, 449, 503, 491, 523, 492, 511, 465, 449, 483, 515, 507, 495, 452, 444, 523, 476, 513, 429, 488, 458, 471, 458, 463, 476, 486, 501, 502, 468, 525, 477, 453, 443, 443, 457, 490, 478, 435, 484, 451, 491, 481, 431, 430, 430, 435, 451, 424, 475, 411, 486, 401, 455, 466, 522, 456, 471, 492, 437, 510, 492, 508, 480, 436, 436, 422, 438, 453, 471, 440, 412, 442, 463, 458, 428, 525, 454, 428, 424, 439, 402, 481, 435, 432, 430, 452, 478, 396, 439, 474, 468, 449, 415, 492, 431, 458, 461, 443, 451, 419, 401, 474, 446, 449, 440, 465, 420, 486, 436, 465, 491, 467, 460, 429, 480, 536, 496, 454, 434, 457, 427, 533, 487, 484, 519, 480, 529, 454, 494, 497, 471, 456, 469, 461, 417, 463, 489, 447, 471, 420, 420, 412, 429, 428, 466, 455, 460, 453, 460, 458, 431, 440, 431, 408, 448, 417, 475, 427, 515, 500, 509, 492, 457, 461, 439, 491, 433, 398, 460, 461, 479, 430, 479, 422, 462, 502, 462, 415, 505, 460, 468, 457, 479, 490, 492, 500, 484, 260, 414, 447, 422, 417, 468, 531, 512, 580, 518, 435, 388, 482, 413, 430, 446, 478, 348, 421, 486, 478, 504, 465, 452, 378, 448, 445, 433, 495, 552, 445, 483, 446, 521, 466, 494, 536, 461, 474, 495, 473, 461, 489, 504, 444, 506, 501, 498, 519, 470, 480, 487, 474, 453, 497, 539, 494, 526, 510, 431, 473, 530, 533, 439, 485, 469, 479, 492, 555, 568, 513, 461, 506, 529, 457, 494, 490, 445, 464, 496, 475, 490, 413, 426, 503, 483, 493, 470, 460, 526, 553, 583, 579, 552, 595, 533, 525, 502, 559, 502, 534, 567, 574, 577, 521, 495]
[117, 75, 118, 132, 104, 93, 118, 155, 140, 110, 131, 118, 140, 114, 141, 108, 136, 130, 151, 136, 129, 141, 175, 124, 155, 167, 116, 152, 115, 144, 129, 136, 125, 170, 137, 131, 135, 161, 154, 165, 156, 135, 98, 110, 95, 128, 158, 116, 137, 137, 166, 110, 149, 180, 157, 163, 154, 146, 113, 138, 163, 143, 133, 153, 141, 156, 144, 167, 138, 161, 175, 170, 122, 150, 176, 146, 116, 141, 130, 138, 121, 148, 112, 299, 136, 98, 147, 115, 125, 143, 167, 146, 119, 129, 127, 127, 139, 146, 128, 118, 139, 138, 125, 137, 109, 126, 102, 216, 126, 129, 122, 115, 126, 131, 102, 114, 109, 139, 123, 142, 145, 114, 116, 139, 123, 126, 139, 113, 74, 96, 110, 131, 130, 140, 130, 135, 140, 121, 130, 109, 114, 139, 124, 136, 114, 134, 95, 121, 125, 139, 122, 145, 136, 146, 132, 139, 162, 145, 112, 121, 88, 96, 117, 117, 127, 102, 102, 114, 134, 152, 92, 103, 142, 134, 129, 129, 126, 147, 115, 149, 133, 188, 99, 122, 151, 110, 137, 110, 127, 136, 136, 118, 146, 130, 134, 97, 127, 149, 134, 134, 102, 158, 137, 144, 171, 155, 145, 151, 135, 148, 157, 124, 143, 147, 167, 145, 156, 129, 198, 140, 146, 129, 123, 143, 150, 136, 143, 146, 148, 166, 129, 117, 111, 119, 108, 137, 124, 127, 138, 142, 167, 131, 150, 135, 147, 160, 152, 169, 133, 184, 171, 137, 159, 136, 128, 148, 127, 137, 125, 196, 164, 133, 155, 177, 153, 163, 143, 145, 68, 99, 110, 130, 121, 134, 147, 125, 138, 170, 149, 160, 124, 146, 120, 175, 169, 143, 122, 129, 108, 126, 168, 109, 128, 141, 151, 137, 126, 55, 102, 127, 155, 154, 135, 124, 75, 106, 135, 150, 152, 140, 162, 168, 154, 117, 214, 152, 118, 108, 101, 126, 108, 136, 121, 134, 136, 119, 93, 110, 111, 115, 103, 124, 82, 122, 117, 89, 105, 107, 99, 121, 121, 127, 141, 87, 108, 131, 87, 137, 132, 107, 121, 119, 110, 94, 143, 102, 133, 143, 138, 110, 151, 132, 138, 125, 125, 148, 102, 112, 124, 119, 103, 140, 105, 122, 152, 167, 194, 152, 141, 165, 149, 157, 125, 195, 178, 148, 158, 112, 152, 131, 136, 108, 143, 130, 136, 138, 169, 166, 106, 128, 128, 153, 175]

# Calculating the percentage of sentiment in each category
neg_sentiment_arr = np.array(negative_sentiment)
pos_sentiment_arr = np.array(positive_sentiment)
total_neg_day = (neg_sentiment_arr/N)*100
total_pos_day = (pos_sentiment_arr/N)*100

# Plotting the Sentiment analysis of each day
index = range(len(total_neg_day))
tick_labels = ['Day-' + str(i) for i in range(1, len(total_neg_day)+1, 10)]
plt.figure(figsize=(15,8))
plt.bar(index, total_neg_day, color='blue', label='Negative Sentiment')
plt.bar(index, total_pos_day, color='red', label='Positive Sentiment')
plt.xlabel('Day')
plt.ylabel('Percentage')
plt.title('Bar Chart of sentiment distribution for each day')
plt.xticks(range(0, len(total_neg_day), 10), tick_labels, rotation=45)
plt.legend(loc='upper center', bbox_to_anchor=(0.5, 1.15), ncol=2)
plt.show()

# Getting the count of samples in each week
week_n = []
a = math.floor(len(N)/7)
b = 0
for i in range(a):
  n = (i+1)*7
  week_n.append(sum(N[b:n]))
  b = n
c = len(N)%7
if c !=0:
  week_n.append(sum(N[-c:]))

# Sentiment analysis for each week
positive_sentiment_week = []
negative_sentiment_week = []
a = 0
n = 0
temp = []

for i in week_n:
    n = n + i
    temp.append(df[a:n].shape[0])
    week_predictions = df['prediction'][a:n]
    count_predictions = week_predictions.value_counts()
    negative_sentiment_week.append(count_predictions.get(0, 0))
    positive_sentiment_week.append(count_predictions.get(1, 0))
    a = n

# Calculating the percentage of sentiment in each category
neg_sentiment_arr = np.array(negative_sentiment_week)
pos_sentiment_arr = np.array(positive_sentiment_week)
total_neg_week = (neg_sentiment_arr/week_n)*100
total_pos_week = (pos_sentiment_arr/week_n)*100

# Plotting the Sentiment analysis weekly
index = [f"Week {i+1}" for i in range(len(total_neg_week))]
plt.figure(figsize=(15,8))
plt.bar(index, total_neg_week, color='blue', label='Negative Sentiment')
plt.bar(index, total_pos_week, color='red', label='Positive Sentiment')
# x = np.arange(len(weeks))
plt.xlabel('Week')
plt.ylabel('Percentage')
plt.xticks(index, rotation=45, ha='right')
plt.title('Bar Chart of sentiment distribution for each week')
plt.legend(loc='upper center', bbox_to_anchor=(0.5, 1.15), ncol=2)
plt.show()

# Getting the count of samples in each month
df['Date'] = pd.to_datetime(df['Date'])
df['month_year'] = df['Date'].dt.to_period('M')
counts = df.groupby('month_year').size()
month_count = []
for month, count in counts.items():
    # print(f"{month.strftime('%B %Y')}: {count} days")
    month_count.append(count)

# Sentiment analysis for each month
positive_sentiment_month = []
negative_sentiment_month = []
a = 0
n = 0
temp = []

for i in month_count:
    n = n + i
    temp.append(df[a:n].shape[0])
    month_predictions = df['prediction'][a:n]
    count_predictions = month_predictions.value_counts()
    negative_sentiment_month.append(count_predictions.get(0, 0))
    positive_sentiment_month.append(count_predictions.get(1, 0))
    a = n

# Calculating the percentage of sentiment in each category
neg_sentiment_arr = np.array(negative_sentiment_month)
pos_sentiment_arr = np.array(positive_sentiment_month)
total_neg_month = (neg_sentiment_arr/month_count)*100
total_pos_month = (pos_sentiment_arr/month_count)*100

# Plotting the Sentiment analysis monthly
index = range(len(total_neg_month))
months = ['Feb-22', 'Mar-22', 'Apr-22', 'May-22', 'Jun-22', 'Jul-22', 'Aug-22', 'Sep-22', 'Oct-22', 'Nov-22', 'Dec-22', 'Jan-23', 'Feb-23', 'Mar-23']
plt.figure(figsize=(15,8))
plt.bar(index, total_neg_month, color='blue', label='Negative Sentiment')
plt.bar(index, total_pos_month, color='red', label='Positive Sentiment')
x = np.arange(len(months))
plt.xlabel('Month')
plt.ylabel('Percentage')
plt.xticks(x, months)
plt.title('Bar Chart of sentiment distribution for each month')
plt.legend(loc='upper center', bbox_to_anchor=(0.5, 1.15), ncol=2)
plt.show()

# Creating a dataframe for plotting the time series
index_value = [137.72, 154.50, 148.28, 151.17, 156.25, 156.70, 157.16, 161.71, 171.75, 168.53, 157.40, 158.32, 156.55, np.nan]
time_series_df = pd.DataFrame({'Date': months, 'Negative': total_neg_month, "Positive": total_pos_month, 'Macro data': index_value})
time_series_df

	Date	Negative	Positive	Macro data
0	Feb-22	81.281139	18.718861	137.72
1	Mar-22	79.813049	20.186951	154.50
2	Apr-22	78.884799	21.115201	148.28
3	May-22	77.950108	22.049892	151.17
4	Jun-22	79.647130	20.352870	156.25
5	Jul-22	78.360171	21.639829	156.70
6	Aug-22	78.233996	21.766004	157.16
7	Sep-22	76.018712	23.981288	161.71
8	Oct-22	76.588185	23.411815	171.75
9	Nov-22	76.434369	23.565631	168.53
10	Dec-22	77.206083	22.793917	157.40
11	Jan-23	80.528754	19.471246	158.32
12	Feb-23	80.169217	19.830783	156.55
13	Mar-23	77.630687	22.369313	NaN

# Time-series plot of positive, negative and macro index
plt.figure(figsize=(15,8))
plt.plot(time_series_df['Date'], time_series_df['Negative'], color='blue', label='Negative Sentiment')
plt.plot(time_series_df['Date'], time_series_df['Positive'], color='red',label='Positive Sentiment')
plt.plot(time_series_df['Date'], time_series_df['Macro data'], color = 'green', label = 'Macro Index')

plt.xlabel('Month')
plt.ylabel('Percentage / Macro Value')
plt.title('Sentiment Distribution Over Time')
plt.xticks(rotation=30)
plt.legend(loc='upper center', bbox_to_anchor=(0.5, 1.15), ncol=2)
plt.show()

# Calculating the percentage change of positive, negative and macro index
Percentage_Change = time_series_df[['Negative', 'Positive', 'Macro data']].pct_change()

# Time-series plot of percentage change in positive, negative and macro index
plt.figure(figsize=(15, 8))
plt.plot(time_series_df['Date'], Percentage_Change['Negative'], color='blue', label='Negative Sentiment')
plt.plot(time_series_df['Date'], Percentage_Change['Positive'], color='red', label='Positive Sentiment')
plt.plot(time_series_df['Date'], Percentage_Change['Macro data'], color='green', label='Macro Index')
plt.xlabel('Month')
plt.ylabel('Percentage Change')
plt.title('Sentiment Distribution Over Time (Percentage Change)')
plt.legend(loc='upper center', bbox_to_anchor=(0.5, 1.15), ncol=2)
plt.show()

# Correlation plot for positive, negative and macro index
corr = time_series_df.corr()
plt.figure(figsize=(10,8))
sns.heatmap(corr, vmin=-1, vmax=1, center= 0, cmap='PiYG', annot=True)
plt.show()

Part 3: Factor and topics identification via Machine Learning

# LDA approach to perform topic modeling
def LDA(df):
    docs = [doc.split() for doc in df['Tweet_processed']]
    dictionary = corpora.Dictionary(docs)
    corpus = [dictionary.doc2bow(doc) for doc in docs]
    lda_model = LdaModel(corpus=corpus, id2word=dictionary, num_topics=20, random_state=42)
    topics = lda_model.show_topics(num_topics=20, num_words=10, formatted=False)
    topic_list = []
    for i, topic in topics:
        topic_words = [word for word, _ in topic]
        topic_array = np.array(topic_words)
        topic_list.append(topic_array)
    plot_terms = pyLDAvis.gensim.prepare(lda_model, corpus, dictionary)
    return topic_list, plot_terms

# Negative twitter topics
twitter_neg_sentiment, plot_neg = LDA(df_ukr_tweet)
# twitter_neg_sentiment

# Terms in negative twitter topics
a = pd.concat({'Topics': pd.DataFrame(twitter_neg_sentiment)})
a

		0	1	2	3	4	5	6	7	8	9
Topics	0	war	ukraine	tank	russia	amp	crimes	ukrainewar	media	propaganda	continues
	1	ukraine	russia	war	people	like	dont	world	amp	putin	want
	2	ukraine	us	president	battle	freedom	democracy	britain	united	zelenskyy	states
	3	bakhmut	ukraine	russian	ukrainian	attack	forces	footage	donetsk	drone	kherson
	4	nato	eu	usa	ukraine	russia	amp	us	europe	countries	turkey
	5	germany	canada	america	ukraine	biden	horrible	russia	war	usa	trump
	6	ukraine	win	terrorist	victory	minister	cnn	together	state	foreign	high
	7	china	ukraine	nato	brutal	russia	us	sweden	finland	ua	join
	8	ukraine	poland	group	combat	march	repo	aid	including	glory	fire
	9	ukraine	destroy	germany	nuclear	front	bombs	armed	russia	france	th
	10	standwithukraine	ukrainewillwin	help	video	free	putin	family	stay	hundreds	hungary
	11	soldiers	ukraine	kyiv	russian	ukrainian	germany	killed	troops	special	kiev
	12	ukraine	c	months	truth	todays	stand	evil	min	australia	taking
	13	ukraine	children	people	war	one	syria	russia	lost	armukrainenow	years
	14	russiaisaterroriststate	standwithukraine	leopard	strike	ukraine	gopro	putinisawarcriminal	daily	flag	drops
	15	ukraine	putin	money	mar	ukrainerussiawar	nazis	send	dead	fuck	follow
	16	ukraine	russia	putin	war	amp	crimea	ammunition	russian	elite	criminal
	17	ukraine	amp	invasion	russias	russia	map	iran	billion	war	un
	18	ukraine	kyiv	russia	tanks	drones	troops	russian	ukrainewar	missiles	military
	19	wagner	ukraine	ukrainerussianwar	destroyed	ukrainewar	ukrainerussiawar	soledar	army	slavaukraini	troops

# Relevant terms for the negative sentiment topic
pyLDAvis.display(plot_neg)

# Topic cloud for negative sentiment
topic_text_combined = ' '.join([' '.join(topic_words) for topic_words in twitter_neg_sentiment])
wordcloud = WordCloud(background_color='white').generate(topic_text_combined)
plt.figure(figsize=(12, 10))
plt.imshow(wordcloud, interpolation="bilinear")
plt.axis("off")
plt.title("Topic cloud for negative sentiment")
plt.show()

# Positive twitter topics
twitter_pos_sentiment, plot_pos = LDA(df_rus_tweet)
# twitter_pos_sentiment

# Terms in positive twitter topics
b = pd.concat({'Topics': pd.DataFrame(twitter_pos_sentiment)})
b

		0	1	2	3	4	5	6	7	8	9
Topics	0	ukraine	russia	usa	nato	china	war	us	eu	amp	uk
	1	ukraine	bakhmut	russia	ukrainewar	russian	ukrainerussiawar	ukrainian	ukrainerussianwar	war	russiaukrainewar
	2	ukraine	russia	wagner	kherson	russian	today	th	day	fighters	black
	3	ukraine	germany	good	russia	leopard	equipment	link	via	eu	india
	4	ukraine	nafo	russia	special	top	operation	warinukraine	united	europeanunion	war
	5	ukraine	russia	war	russian	donbas	news	russias	ukrainewar	ukrainerussiawar	putin
	6	ukraine	russian	russia	ukrainerussiawar	drone	ukrainian	footage	ukrainewar	forces	donetsk
	7	ukraine	nazis	well	nazi	life	naziukraine	winter	winning	ukrainerussiawar	ukrainian
	8	video	ukraine	please	zelenskywarcriminal	check	nowar	ukrainewarvideos	ukraina	tiktok	th
	9	ukraine	russia	putin	ukrainerussiawar	russiaisaterroriststate	ukrainewar	war	russian	russiaisateroriststate	russianwarcrimes
	10	ukraine	ukrainewillwin	suppo	ukrainerussiawar	thanks	year	war	ukrainewar	great	drones
	11	america	ukraine	weapons	vs	russia	soldier	z	turkey	ukrainewar	st
	12	ukraine	ukrainerussiawar	ukrainewar	war	russia	ukrainewarnews	new	ukraineunderattack	nafofellas	russias
	13	canada	ukraine	germany	america	russia	tank	aillery	christmas	fire	gt
	14	ukraine	britain	russia	god	victory	iran	new	istandwithrussia	group	bless
	15	ukraine	russia	free	war	fight	via	team	update	ukrainian	security
	16	ukraine	get	much	make	russia	resist	im	better	money	want
	17	ukraine	standwithukraine	love	good	thank	standwithrussia	music	stopwar	night	amp
	18	ukraine	happy	full	front	day	visit	ukrainewar	nice	forget	list
	19	ukraine	like	ukrainenews	recipe	looks	slavaukrani	afu	putin	full	post

# Relevant terms for the positive sentiment topic
pyLDAvis.display(plot_pos)

# Topic cloud for positive sentiment
topic_text_combined = ' '.join([' '.join(topic_words) for topic_words in twitter_pos_sentiment])
wordcloud = WordCloud(background_color='white').generate(topic_text_combined)
plt.figure(figsize=(12, 10))
plt.imshow(wordcloud, interpolation="bilinear")
plt.axis("off")
plt.title("Topic cloud for positive sentiment")
plt.show()

# Random 50 sample for negative sentiment
df_neg_tweet=pd.DataFrame(np.random.permutation(df[idx_ukr]),columns=df.columns)

df_neg_sample=df_neg_tweet.iloc[0:50,:]
df_neg_sample.to_csv('twitter_negative_sample.csv', sep=',', index=False)
df_neg_sample

	Datetime	Username	Text	Language	Location	Coordinates	Likes	Retweets	Hashtags	URL	Tweet	Date	prediction	positive_sentiment	negative_sentiment	month_year
0	2022-10-26 22:29:40+00:00	OneVenusThrow	['kill', 'everyonerussian', 'violenc', 'ukrain...	en	Manna-hata	NaN	0	1	['Ukraine', 'brutality', 'WarCrimes', 'Russia']	https://twitter.com/OneVenusThrow/status/15853...	'Kill everyone':Russian violence in #Ukraine w...	2022-10-26	0	0	1	2022-10
1	2022-08-05 22:35:18+00:00	SanctionsAml	['biden', 'elect', 'trump', 'foreign', 'polici...	en	NaN	NaN	1	1	['Biden', 'Trump', 'Taliban', 'Iran', 'Putin',...	https://twitter.com/SanctionsAml/status/155568...	#Biden - elected for "not being #Trump". But h...	2022-08-05	0	0	1	2022-08
2	2022-07-18 23:29:53+00:00	theghostgiver	['usip', 'vladimir', 'putin', 'may', 'limit', ...	en	California, USA	NaN	0	0	['Ukraine']	https://twitter.com/theghostgiver/status/15491...	USIP: While Vladimir Putin may have limited hi...	2022-07-18	0	0	1	2022-07
3	2023-01-29 23:20:48+00:00	Hi45733134	['gopro', 'footag', 'ukrain', 'brutal', 'attac...	en	NaN	NaN	0	0	['Canada', 'Germany', 'America', 'Ukraine', 'B...	https://twitter.com/Hi45733134/status/16198379...	GoPro footage! Ukraine's Brutal Attack on Luha...	2023-01-29	0	0	1	2023-01
4	2023-03-29 19:28:47+00:00	RabiaSalem02	['horribl', 'headcam', 'ukrain', 'war', 'video...	en	Libya	NaN	0	0	['Canada', 'Germany', 'America', 'Ukraine', 'B...	https://twitter.com/RabiaSalem02/status/164116...	Horrible head-cam, Ukraine war video footage, ...	2023-03-29	0	0	1	2023-03
5	2022-04-21 23:48:35+00:00	sgervay	['eah', 'day', 'day', 'rememb', 'act', 'save',...	en	Sydney	NaN	1	0	['standwithukraine']	https://twitter.com/sgervay/status/15172892228...	Earth day is the day we remember to act to sav...	2022-04-21	0	0	1	2022-04
6	2022-05-11 23:56:46+00:00	MimiTexasAngel	['cannabi', 'replac', 'fossil', 'fuel', 'eat',...	en	Planet Earth	NaN	0	0	['Cannabis', 'Ukraine']	https://twitter.com/MimiTexasAngel/status/1524...	@SenSherrodBrown @GrandmaShark18 #Cannabis can...	2022-05-11	0	0	1	2022-05
7	2022-02-21 23:56:17+00:00	MateuszSobiera3	['barack', 'obama', 'sinc', 'main', 'threat', ...	en	NaN	NaN	0	0	['BreakingNews', 'Russia', 'Ukraine', 'Poland'...	https://twitter.com/MateuszSobiera3/status/149...	Barack Obama: 'Since the 1980s the main threat...	2022-02-21	0	0	1	2022-02
8	2022-06-17 23:01:38+00:00	art24karat	['agre', 'made', 'mess', 'whole', 'world', 'po...	en	Hopkinton, MA	NaN	0	0	['Ukraine', 'Russia']	https://twitter.com/art24karat/status/15379335...	@rishibagree I agree he has made a mess in the...	2022-06-17	0	0	1	2022-06
9	2022-04-05 23:57:55+00:00	hamidashraf1	['your', 'democrat', 'ukrain', 'dare', 'critic...	en	Iran	NaN	0	0	['ukraine', 'torture']	https://twitter.com/hamidashraf1/status/151149...	@Soheil34874913 If you're in "democratic" #ukr...	2022-04-05	0	0	1	2022-04
10	2022-05-25 23:32:06+00:00	PeggerCAP	['gun', 'owner', 'unit', 'amp', 'pack', 'milit...	en	Canada	NaN	9	2	['Ukraine', 'SlavaUkraini']	https://twitter.com/PeggerCAP/status/152960626...	@TheLeadCNN @jaketapper Gun owners should unit...	2022-05-25	0	0	1	2022-05
11	2022-12-07 22:43:47+00:00	ajonestime63	['right', 'mani', 'peac', 'deal', 'afghanistan...	en	NaN	NaN	0	0	['Afghanistan', 'Ukraine', 'Russia', 'NorthKor...	https://twitter.com/ajonestime63/status/160062...	@Dr_Bry_So_Fly Right… how many peace deals did...	2022-12-07	0	0	1	2022-12
12	2022-04-24 23:10:51+00:00	SanctionsAhoy	['russia', 'oligarch', 'aleksandr', 'frolov', ...	en	NaN	NaN	0	0	['Russia', 'oligarch', 'Frolov', 'yacht', 'Sta...	https://twitter.com/SanctionsAhoy/status/15183...	#Russia #oligarch Aleksandr #Frolov's #yacht D...	2022-04-24	0	0	1	2022-04
13	2022-05-25 23:37:39+00:00	Turkwoyss	['urgent', 'appeal', 'lessen', 'pet', 'peopl',...	en	NaN	NaN	0	0	['Urgent', 'appeal', 'pet', 'people', 'pets', ...	https://twitter.com/Turkwoyss/status/152960766...	#Urgent #appeal: Lessen #pet paperwork/cost fo...	2022-05-25	0	0	1	2022-05
14	2022-03-18 23:28:50+00:00	MikeMillburnTV	['come', 'pm', 'introduc', 'mother', 'daughter...	en	San Diego, CA	NaN	3	2	['Ukraine', 'KUSINews']	https://twitter.com/MikeMillburnTV/status/1504...	Coming up on the @KUSINews at 5pm:\n\n@MattPri...	2022-03-18	0	0	1	2022-03
15	2022-09-12 23:09:17+00:00	SGNewsAlerts	['zelenski', 'night', 'appeal', 'stoprussianag...	en	NaN	NaN	1	0	['StopRussianAgression', 'StopRussia', 'Ukrain...	https://twitter.com/SGNewsAlerts/status/156946...	Zelensky’s night appeal: \n#StopRussianAgressi...	2022-09-12	0	0	1	2022-09
16	2022-09-24 23:50:56+00:00	OneVenusThrow	['histor', 'collaps', 'journal', 'patricklawr'...	en	Manna-hata	NaN	0	0	['journalism', 'PatrickLawrence', 'Ukraine', '...	https://twitter.com/OneVenusThrow/status/15738...	"The historic collapse of #journalism"\nBy #Pa...	2022-09-24	0	0	1	2022-09
17	2022-09-24 22:34:05+00:00	apmozaffari	['give', 'us', 'time', 'liber', 'iran', 'mulla...	en	NaN	NaN	0	0	['Iran', 'Ukraine', 'MahsaAmini']	https://twitter.com/apmozaffari/status/1573802...	@apmassaro3 If you give us some time, we will ...	2022-09-24	0	0	1	2022-09
18	2022-10-16 22:09:54+00:00	PeterClifford1	['ukrain', 'ua', 'gener', 'staff', 'repo', 'ru...	en	UK/FR	NaN	3	1	['Ukraine', 'Kherson', 'Crimea', 'tweet100']	https://twitter.com/PeterClifford1/status/1581...	#Ukraine: 4. UA’s General Staff reports that R...	2022-10-16	0	0	1	2022-10
19	2022-03-01 23:51:16+00:00	i24NEWS_EN	['flee', 'ukrain', 'russian', 'invas', 'goe', ...	en	Jaffa Port	NaN	4	5	['Ukraine']	https://twitter.com/i24NEWS_EN/status/14988081...	More than 600,000 flee #Ukraine as the Russian...	2022-03-01	0	0	1	2022-03
20	2022-12-25 22:52:42+00:00	Palcw	['merri', 'christma', 'whole', 'western', 'wor...	en	Bergen, Norway	NaN	2	0	['StandWithUkraine']	https://twitter.com/Palcw/status/1607147349733...	@Kateryna_Kruk Merry Christmas! The whole West...	2022-12-25	0	0	1	2022-12
21	2022-09-23 22:23:55+00:00	DJFunkyGrrL	['russia', 'vladimir', 'putin', 'mobil', 'puti...	en	Travel: Sicily, Sweden, U.S	NaN	0	0	['Politics', 'News', 'Russia', 'Putin', 'Ukrai...	https://twitter.com/DJFunkyGrrL/status/1573438...	[ Russia ]\n.\n.. . about Vladimir Putin’s mob...	2022-09-23	0	0	1	2022-09
22	2022-10-16 22:30:21+00:00	milposp	['kind', 'suppo', 'victori', 'ukrain', 'assur'...	en	NaN	NaN	0	0	['Ukraine']	https://twitter.com/milposp/status/15817745752...	With this kind of support, victory for #Ukrain...	2022-10-16	0	0	1	2022-10
23	2022-03-07 23:48:43+00:00	Hope_Resists	['standwithukrain', 'stopputinnow', 'zelenskiy...	en	NaN	NaN	0	0	['StandWithUkraine', 'StopPutinNOW', 'Zelenskiy']	https://twitter.com/Hope_Resists/status/150098...	#StandWithUkraine #StopPutinNOW 03/07/2022 #Ze...	2022-03-07	0	0	1	2022-03
24	2022-09-09 23:59:02+00:00	Cicke69	['highli', 'contextu', 'question', 'depend', '...	en	Belgrade City	NaN	5	1	['ukrainerussiawar', 'kharkov']	https://twitter.com/Cicke69/status/15683885447...	This is a highly contextual question, which de...	2022-09-09	0	0	1	2022-09
25	2022-12-07 22:37:03+00:00	HiGk2k	['ukrain', 'war', 'fight', 'continu', 'bakhmut...	en	NaN	NaN	0	0	['Canada', 'Germany', 'America', 'Ukraine', 'B...	https://twitter.com/HiGk2k/status/160062042956...	Ukraine War: Fighting continues in Bakhmut\n#C...	2022-12-07	0	0	1	2022-12
26	2022-11-23 22:21:55+00:00	OksanaLviv2	['russia', 'recogn', 'terrorist', 'state', 'wo...	en	NaN	NaN	2	0	['Russia', 'Kuleba', 'Ukraine', 'russiaisatero...	https://twitter.com/OksanaLviv2/status/1595543...	"#Russia should be recognized as a terrorist s...	2022-11-23	0	0	1	2022-11
27	2022-12-06 22:38:08+00:00	Kabmin_UA_e	['ukrain', 'energi', 'system', 'remain', 'inte...	en	Kyiv, Ukraine	NaN	8	0	['EnergySecurity', 'StandWithUkraine', 'Emerge...	https://twitter.com/Kabmin_UA_e/status/1600258...	Ukraine’s energy system remains integral and u...	2022-12-06	0	0	1	2022-12
28	2022-11-08 22:35:05+00:00	LordOfParadiseU	['boom', 'ukrain', 'russia', 'nato', 'eu', 'po...	en	somewhere in Heaven	NaN	1	1	['ukraine', 'russia', 'NATO', 'EU', 'potus', '...	https://twitter.com/LordOfParadiseU/status/159...	BOOM.\n#ukraine #russia #NATO #EU #potus #kyiv...	2022-11-08	0	0	1	2022-11
29	2022-05-27 22:19:20+00:00	RedTitans3	['europ', 'readi', 'give', 'putin', 'piec', 'l...	en	NaN	NaN	0	0	['UkraineWar', 'UkraineUnderAttaсk', 'UnitedWi...	https://twitter.com/RedTitans3/status/15303127...	Just a thought.Would Europes be so ready to gi...	2022-05-27	0	0	1	2022-05
30	2022-07-01 23:23:07+00:00	ShehabBawazeer2	['still', 'believ', 'nato', 'innoc', 'stupid',...	en	NaN	NaN	28	16	['Russia', 'Ukraine️', 'Russian', 'Ukrainian',...	https://twitter.com/ShehabBawazeer2/status/154...	If you still believe in NATOs innocence you ar...	2022-07-01	0	0	1	2022-07
31	2022-10-13 22:29:24+00:00	Its_Carlos_Here	['ban', 'confeder', 'symbol', 'statu', 'battl'...	en	USA - Mexico	NaN	0	0	['Ukraine', 'Putin', 'Biden']	https://twitter.com/Its_Carlos_Here/status/158...	If they ban confederate symbols, statues and t...	2022-10-13	0	0	1	2022-10
32	2023-02-15 22:56:24+00:00	Husla3xGENIUS	['everyon', 'wont', 'see', 'song', 'song', 'de...	en	Wisconsin, USA	Coordinates(longitude=-88.070827, latitude=42....	2	1	['UkraineWar', 'loveisland', 'fearwomen']	https://twitter.com/Husla3xGENIUS/status/16259...	Everyone won't see now in this song but this s...	2023-02-15	0	0	1	2023-02
33	2022-10-12 23:41:04+00:00	NoMofreedomm	['type', 'handshak', 'give', 'boy', 'bang', 'r...	en	NaN	NaN	0	0	['UkraineRussianWar', 'Ukraine', 'Kiev']	https://twitter.com/NoMofreedomm/status/158034...	This is the type of handshake you give your bo...	2022-10-12	0	0	1	2022-10
34	2022-12-30 22:23:41+00:00	arieladkins	['dennisrodman', 'hungri', 'threw', 'black', '...	en	California, USA	NaN	0	0	['DennisRodman', 'CharlesOakley', 'greed', 'ka...	https://twitter.com/arieladkins/status/1608951...	#DennisRodman was hungry and #CharlesOakley th...	2022-12-30	0	0	1	2022-12
35	2023-02-06 22:15:08+00:00	September_UA	['took', 'filtrat', 'camp', 'separ', 'mother',...	en	Canada	NaN	0	0	['Ukraine', 'USA', 'Germany', 'EU', 'UK', 'Eur...	https://twitter.com/September_UA/status/162272...	They took me to a filtration camp and separate...	2023-02-06	0	0	1	2023-02
36	2022-11-19 23:57:16+00:00	MavrikM221	['forget', 'extern', 'influenc', 'usa', 'ukrai...	en	United Kingdom	NaN	0	0	['USA', 'UkraineRussiaWar', 'China', 'NorthKor...	https://twitter.com/MavrikM221/status/15941176...	Forget not the external influences to #USA - ...	2022-11-19	0	0	1	2022-11
37	2022-11-08 22:22:11+00:00	RimlandSky	['plan', 'annex', 'belgorod', 'talk', 'arm', '...	en	Old Europe (NATO/EU)	NaN	0	0	['NATO', 'Ukraine']	https://twitter.com/RimlandSky/status/15901074...	Not planning to annex Belgorod and talking abo...	2022-11-08	0	0	1	2022-11
38	2023-02-10 23:26:38+00:00	HiHikeep4	['massiv', 'attack', 'ukrainian', 'forc', 'fli...	en	Kentucky, USA	NaN	1	0	['Canada', 'Germany', 'America', 'Ukraine', 'B...	https://twitter.com/HiHikeep4/status/162418811...	Massive Attack!!! Ukrainian forces fly Drones ...	2023-02-10	0	0	1	2023-02
39	2023-03-29 19:33:12+00:00	Hi45733134	['gopro', 'footag', 'ukrainian', 'arm', 'forc'...	en	NaN	NaN	0	0	['Canada', 'Germany', 'America', 'Ukraine', 'B...	https://twitter.com/Hi45733134/status/16411616...	GoPro Footage !! The Ukrainian Armed Forces ov...	2023-03-29	0	0	1	2023-03
40	2023-02-23 23:20:44+00:00	trade_yyc	['deliber', 'derail', 'train', 'kill', 'five',...	en	NaN	NaN	11	7	['derailed', 'environmentalscientists', 'Ukrai...	https://twitter.com/trade_yyc/status/162889767...	They deliberately #derailed the train then kil...	2023-02-23	0	0	1	2023-02
41	2022-09-10 22:23:28+00:00	jennifer_jj	['fluke', 'post', 'came', 'backtoback', 'timel...	en	Nomad (St Thomas, ON for now)	NaN	1	0	['Ukraine']	https://twitter.com/jennifer_jj/status/1568726...	Fluke of RTs, these posts from @danrather and ...	2022-09-10	0	0	1	2022-09
42	2022-03-31 23:15:15+00:00	HeineGiessen	['stand', 'ukrain', 'ukrain', 'ukrainian', 'na...	en	European Union: Frankfurt/Main	NaN	0	0	['Ukraine', 'EU']	https://twitter.com/HeineGiessen/status/150967...	Stand With Ukraine #Ukraine 🇺🇦\nUkrainian Nati...	2022-03-31	0	0	1	2022-03
43	2023-02-14 22:58:34+00:00	1988Letzfetz	['nato', 'shame']	en	NaN	NaN	0	0	['Nato']	https://twitter.com/1988Letzfetz/status/162563...	@ukrainiansquad #Nato is a shame	2023-02-14	0	0	1	2023-02
44	2022-07-14 23:43:08+00:00	Ohforfuckssake2	['pleas', 'pleas', 'standwithukrain']	en	Remote Biolab Field Office	NaN	0	0	['StandWithUkraine']	https://twitter.com/Ohforfuckssake2/status/154...	@POTUS @KamalaHarris Please, please #StandWith...	2022-07-14	0	0	1	2022-07
45	2022-04-19 23:41:00+00:00	ManoDiTerzaroli	['becom', 'orc', 'mommi', 'instead', 'wash', '...	en	Constantinople	NaN	0	0	['StandWithUkraine', 'WarCrimes', 'StopRussian...	https://twitter.com/ManoDiTerzaroli/status/151...	How to become an Orc: when mommy instead of sh...	2022-04-19	0	0	1	2022-04
46	2022-05-22 22:54:03+00:00	TarasDudykevych	['german', 'mp', 'told', 'german', 'tv', 'does...	en	NaN	NaN	1	1	['Ukraine']	https://twitter.com/TarasDudykevych/status/152...	@jjroach3 @AVindman @ZelenskyyUa German MP @RK...	2022-05-22	0	0	1	2022-05
47	2022-04-11 23:03:09+00:00	NoLiberals	['bodi', 'discov', 'mass', 'grave', 'town', 'v...	en	Boston, MA	NaN	0	0	['news', 'Ukraine']	https://twitter.com/NoLiberals/status/15136539...	More than 1,200 bodies have now been discovere...	2022-04-11	0	0	1	2022-04
48	2022-04-22 23:02:51+00:00	AnonymaticaSRL	['shell', 'mine', 'found', 'soviet', 'stock', ...	en	Tettenweis, Deutschland	NaN	2	0	['UkraineRussiaWar', 'RussiaUkraineWar', 'Ukra...	https://twitter.com/AnonymaticaSRL/status/1517...	@Navsteva @Lukyluke311 “Shells and mines were ...	2022-04-22	0	0	1	2022-04
49	2022-03-13 23:44:23+00:00	BuddhaThehun	['sound', 'come', 'goe', 'ukrain']	en	The Medium Place	NaN	0	0	['Ukraine']	https://twitter.com/BuddhaThehun/status/150315...	Sounds comes and goes #Ukraine https://t.co/h...	2022-03-13	0	0	1	2022-03

# Random 50 sample for positive sentiment
df_pos_tweet=pd.DataFrame(np.random.permutation(df[idx_rus]),columns=df.columns)

df_pos_sample=df_pos_tweet.iloc[0:50,:]
df_pos_sample.to_csv('twitter_positive_sample.csv', sep=',', index=False)
df_pos_sample

	Datetime	Username	Text	Language	Location	...	Date	prediction	positive_sentiment	negative_sentiment	month_year
0	2022-05-31 22:46:00+00:00	JFK_ALlivestill	['bunch', 'rot', 'real', 'truth', 'unlik', 'uk...	en	Atlanta, GA, USA	...	2022-05-31	1	1	0	2022-05
1	2022-09-20 23:27:01+00:00	knittingknots	['thread', 'ukrain', 'ukrainerussiawar', 'ukra...	en	in the Minnesota North Woods	...	2022-09-20	1	1	0	2022-09
2	2022-11-13 23:49:35+00:00	AlexMalejewicz	['theyr', 'call', 'ukrainian', 'havent', 'got'...	en	新潟市日本	...	2022-11-13	1	1	0	2022-11
3	2023-01-22 22:47:23+00:00	TruthPuke	['armedforc', 'ukrain', 'conceal', 'tank', 'ar...	en	Calabasas, CA	...	2023-01-22	1	1	0	2023-01
4	2022-12-28 22:17:26+00:00	ReportWar1	['new', 'footag', 'kharkov', 'explos', 'facil'...	en	NaN	...	2022-12-28	1	1	0	2022-12
5	2022-11-14 22:40:00+00:00	24talker	['cia', 'director', 'meet', 'russian', 'counte...	en	United States	...	2022-11-14	1	1	0	2022-11
6	2022-10-21 22:01:57+00:00	DesignMobyy	['stand', 'ukrain', 'tshi', 'link', 'kiev', 'u...	en	NaN	...	2022-10-21	1	1	0	2022-10
7	2022-03-04 23:56:03+00:00	OneVenusThrow	['latest', 'there', 'standoff', 'ukrainian', '...	en	Manna-hata	...	2022-03-04	1	1	0	2022-03
8	2023-02-28 23:35:01+00:00	david_n919	['battl', 'bakhmut', 'war', 'germani', 'ukrain...	en	NaN	...	2023-02-28	1	1	0	2023-02
9	2022-05-23 22:49:54+00:00	katusha4145	['ukrainian', 'want', 'pa', 'russia', 'bbc', '...	en	Behind the enemy lines.	...	2022-05-23	1	1	0	2022-05
10	2023-03-26 23:26:32+00:00	najy_221	['war', 'ukrain', 'new', 'allianc', 'divid', '...	en	NaN	...	2023-03-26	1	1	0	2023-03
11	2022-07-22 23:30:26+00:00	OKCityRyan	['follow', 'reliabl', 'analysi', 'latest', 'de...	en	Howdy Arabia, USA	...	2022-07-22	1	1	0	2022-07
12	2022-10-12 23:57:13+00:00	akatsuk45629101	['resolut', 'curb', 'desir', 'putin', 'ukrain'...	en	NaN	...	2022-10-12	1	1	0	2022-10
13	2022-06-08 22:17:42+00:00	jason78proton	['russia', 'run', 'tank', 'accord', 'mani', 'm...	en	Philadelphia, PA, USA	...	2022-06-08	1	1	0	2022-06
14	2022-11-09 22:55:29+00:00	sarasasa86	['crimea', 'ukrain', 'temporarili', 'occupi', ...	en	Japan 関西🇯🇵	...	2022-11-09	1	1	0	2022-11
15	2022-03-07 23:38:59+00:00	Vatsek	['anoth', 'one', 'bite', 'dust', 'ukrain']	en	Santa Clara, CA	...	2022-03-07	1	1	0	2022-03
16	2022-07-18 23:15:58+00:00	JanieResists	['russian', 'ukrainian', 'ukrainewar', 'russia...	en	DC USA: I buy @BonaFideMasks	...	2022-07-18	1	1	0	2022-07
17	2022-08-18 23:13:04+00:00	knittingknots	['russian', 'militari', 'vehicl', 'insid', 'se...	en	in the Minnesota North Woods	...	2022-08-18	1	1	0	2022-08
18	2022-11-05 23:03:41+00:00	uasupport999	['russia', 'yet', 'destroy', 'singl', 'himar',...	en	Interwebs	...	2022-11-05	1	1	0	2022-11
19	2022-04-13 23:58:24+00:00	knittingknots	['cant', 'fix', 'shatter', 'irpin', 'sign', 'h...	en	in the Minnesota North Woods	...	2022-04-13	1	1	0	2022-04
20	2022-09-24 22:31:31+00:00	Thewarzone7	['oabr', 'name', 'grand', 'duke', 'vytauta', '...	en	NaN	...	2022-09-24	1	1	0	2022-09
21	2022-11-05 21:58:51+00:00	BlogUkraine	['dayofwar', 'arm', 'forc', 'ukrain', 'destroy...	en	Ukraine	...	2022-11-05	1	1	0	2022-11
22	2022-12-29 22:45:41+00:00	nigroeneveld	['new', 'york', 'time', 'investig', 'walk', 'p...	en	The Netherlands	...	2022-12-29	1	1	0	2022-12
23	2022-10-27 23:21:23+00:00	pngtug21	['putin', 'watch', 'russian', 'nuclear', 'dril...	en	NaN	...	2022-10-27	1	1	0	2022-10
24	2023-02-01 23:19:02+00:00	dee_le_brun	['chatgpt', 'told', 'uk', 'agricultur', 'amp',...	en	Sydney	...	2023-02-01	1	1	0	2023-02
25	2022-05-26 23:58:11+00:00	ArvadaRadio	['standwithukrain', 'play', 'dokken', 'love', ...	en	Colorado, USA	...	2022-05-26	1	1	0	2022-05
26	2022-03-25 23:48:27+00:00	NewsAboutTurke2	['serbest', 'gr', 'accord', 'turkey', 'erdogan...	en	Predicting the unpredictable!	...	2022-03-25	1	1	0	2022-03
27	2022-06-05 22:46:42+00:00	sfedorowich	['rate', 'attrit', 'russian', 'brass', 'pretti...	en	Winnipeg Canada	...	2022-06-05	1	1	0	2022-06
28	2023-03-02 22:33:31+00:00	RPerrocalentero	['proukrainian', 'insurg', 'group', 'crimea', ...	en	Telegram	...	2023-03-02	1	1	0	2023-03
29	2022-07-01 22:24:41+00:00	AugustCohen4	['day', 'russia', 'socal', 'great', 'power', '...	en	New York, NY	...	2022-07-01	1	1	0	2022-07
30	2023-01-07 21:58:26+00:00	GroundHoosier	['ukrainewar', 'mccahi', 'use', 'us', 'piggi',...	en	NaN	...	2023-01-07	1	1	0	2023-01
31	2023-02-24 23:49:23+00:00	zeekeron	['differ', 'war', 'term', 'us', 'interest']	en	Global Citizen	...	2023-02-24	1	1	0	2023-02
32	2022-10-30 22:57:24+00:00	Monger_co	['necess', 'mother', 'invent', 'ukrainerussiaw...	en	Washington, DC	...	2022-10-30	1	1	0	2022-10
33	2022-08-05 23:16:01+00:00	IntoxilyzerJ	['war', 'war', 'never', 'chang', 'ukrainewar',...	en	the roadside	...	2022-08-05	1	1	0	2022-08
34	2022-09-17 22:11:06+00:00	europeanMedia24	['kateryna', 'tyshchenko', 'saturday', 'septem...	en	London	...	2022-09-17	1	1	0	2022-09
35	2022-04-28 22:59:15+00:00	wasanthapw	['impact', 'sanction', 'russia', 'ukrain', 'pu...	en	Toronto, Ontario	...	2022-04-28	1	1	0	2022-04
36	2023-03-02 22:28:12+00:00	Suspender_Sue	['possibl', 'chines', 'deal', 'end', 'war', 'r...	en	NaN	...	2023-03-02	1	1	0	2023-03
37	2022-10-08 23:59:01+00:00	Zabluddyh	['russia', 'terrori', 'russia', 'fire', 'missi...	en	NaN	...	2022-10-08	1	1	0	2022-10
38	2022-08-21 22:11:01+00:00	HeineGiessen	['good', 'night', 'eu', 'germani', 'ukrain', '...	en	European Union: Frankfurt/Main	...	2022-08-21	1	1	0	2022-08
39	2022-12-07 23:24:33+00:00	far_east_news	['ukrain', 'war', 'footag', 'ukrainian', 'spec...	en	NaN	...	2022-12-07	1	1	0	2022-12
40	2022-12-26 23:20:54+00:00	luvyall22	['dmitri', 'medvedev', 'new', 'year', 'eve', '...	en	EuropeNorth	...	2022-12-26	1	1	0	2022-12
41	2022-03-10 23:41:54+00:00	aidler	['come', 'seiz', 'boycottrussia', 'ukrainewar']	en	NaN	...	2022-03-10	1	1	0	2022-03
42	2023-03-09 22:22:16+00:00	UkrainianMusic2	['meet', 'ump', 'musician', 'yuliia', 'shevche...	en	NaN	...	2023-03-09	1	1	0	2023-03
43	2022-12-23 22:23:05+00:00	FanduelProTim	['guy', 'actor', 'sinc', 'ukrain', 'becam', 's...	en	United States	...	2022-12-23	1	1	0	2022-12
44	2023-01-02 22:11:38+00:00	donoghue_patty	['check', 'canva', 'print', 'ukrain', 'interio...	en	North Carolina, USA	...	2023-01-02	1	1	0	2023-01
45	2022-03-03 23:53:23+00:00	AkhalTekeFound	['hadrian', 'borysthen', 'akhaltek', 'ancient'...	en	Nationwide	...	2022-03-03	1	1	0	2022-03
46	2023-03-29 18:25:45+00:00	salah_shoaib	['real', 'agenda', 'behind', 'biden', 'proxi',...	en	Dhaka, Bangladesh	...	2023-03-29	1	1	0	2023-03
47	2022-06-11 22:19:35+00:00	Beetman1963	['russian', 'version', 'infam', 'russianwarcri...	en	Tennessee, USA	...	2022-06-11	1	1	0	2022-06
48	2023-03-15 22:31:36+00:00	BeyondSkin	['make', 'music', 'amp', 'danc', 'intern', 'fo...	en	Northern Ireland	...	2023-03-15	1	1	0	2023-03
49	2022-07-18 23:49:48+00:00	radezz	['shit', 'russia', 'ukrain']	en	Belgrade/Serbia/Europe/Earth	...	2022-07-18	1	1	0	2022-07

50 rows × 16 columns

# Correlation of likes and retweets with prediction
correlation = df[['prediction', 'Likes', 'Retweets']].corr()
correlation
plt.figure(figsize=(10,8))
sns.heatmap(correlation, vmin=-1, vmax=1, center= 0, cmap=sns.color_palette("Blues", as_cmap=True), annot=True)
plt.show()

# Correlation of likes and retweets with positive and negative sentiment
correlation = df[['positive_sentiment', 'negative_sentiment', 'Likes', 'Retweets']].corr()
correlation
plt.figure(figsize=(10,8))
sns.heatmap(correlation, vmin=-1, vmax=1, center= 0, cmap=sns.color_palette("Blues", as_cmap=True), annot=True)
plt.show()

Sentiment analysis for specific dates (and one day after) with respect to the important events of the war

# Function that gets the count of each prediction for a single day
def sentiment_analysis(day, df):
  n = df['Date'].value_counts()[str(day)]
  df['Date'] = pd.to_datetime(df['Date'])
  date = pd.to_datetime(str(day))
  df_day = df[df['Date'] == date]
  count_ens_model_day = collections.Counter(df_day['prediction'])
  return count_ens_model_day

# Function that plots pie chart of sentiment analysis for three days (previous, current, and next day) 
def pie_plot(count_ens_model_prev_day, count_ens_model_day, count_ens_model_next_day, a, b, c):
  fig, axes = plt.subplots(nrows=1, ncols=3, figsize=(12, 8))
  ax1 = axes[0]
  ax2 = axes[1]
  ax3 = axes[2]  

  total_1 = count_ens_model_prev_day[0] + count_ens_model_prev_day[1]
  total_2 = count_ens_model_day[0] + count_ens_model_day[1]
  total_3 = count_ens_model_next_day[0] + count_ens_model_next_day[1]

  ax1.pie([count_ens_model_prev_day[0],count_ens_model_prev_day[1]],labels=['{:.2f}%'.format(count_ens_model_prev_day[0]/total_1*100),'{:.2f}%'.format(count_ens_model_prev_day[1]/total_1*100)])
  ax2.pie([count_ens_model_day[0],count_ens_model_day[1]],labels=['{:.2f}%'.format(count_ens_model_day[0]/total_2*100),'{:.2f}%'.format(count_ens_model_day[1]/total_2*100)])
  ax3.pie([count_ens_model_next_day[0],count_ens_model_next_day[1]],labels=['{:.2f}%'.format(count_ens_model_next_day[0]/total_3*100),'{:.2f}%'.format(count_ens_model_next_day[1]/total_3*100)])

  circle_1 = plt.Circle((0, 0), 0.7, color='white')
  circle_2 = plt.Circle((0, 0), 0.7, color='white')
  circle_3 = plt.Circle((0, 0), 0.7, color='white')
  axes[0].add_artist(circle_1)
  axes[1].add_artist(circle_2)
  axes[2].add_artist(circle_3)

  axes[0].set_title('{}'.format(a))
  axes[1].set_title('{}'.format(b))
  axes[2].set_title('{}'.format(c))

  fig.suptitle('Sentiment analysis for Previous, Current, and Next Day', fontsize=14, y=0.75, fontweight='bold')

  ax3.legend(['Negative Sentiment', 'Positive Sentiment'], loc='upper right', bbox_to_anchor=(2, 1))
  plt.show()

# This function performs topic modelling using LDA on the given dataframe df
def LDA_Event(df, n):   # n is the number of topics 
    docs = [doc.split() for doc in df['Tweet_processed']]
    dictionary = corpora.Dictionary(docs)
    corpus = [dictionary.doc2bow(doc) for doc in docs]
    lda_model = LdaModel(corpus=corpus, id2word=dictionary, num_topics=n, random_state=42)
    topics = lda_model.show_topics(num_topics=n, num_words=10, formatted=False)
    topic_list = []
    for i, topic in topics:
        topic_words = [word for word, _ in topic]
        topic_array = np.array(topic_words)
        topic_list.append(topic_array)
    plot_terms = pyLDAvis.gensim.prepare(lda_model, corpus, dictionary)
    return topic_list, plot_terms

March 9th, 2022 - Mariupol maternity hospital attack

# March 8th, 9th and 10th
prev_day = datetime.date(2022, 3, 8)
day = datetime.date(2022, 3, 9)
next_day = datetime.date(2022, 3, 10)
count_ens_model_prev_day = sentiment_analysis(prev_day, df)
count_ens_model_day = sentiment_analysis(day, df)
count_ens_model_next_day = sentiment_analysis(next_day, df)

# Plotting the pie chat
pie_plot(count_ens_model_prev_day, count_ens_model_day, count_ens_model_next_day, 'March 8th', 'March 9th', 'March 10th')

# Relevant terms
date = next_day.strftime('%Y-%m-%d')
twitter_sentiment_event, plot_event = LDA_Event(df_word_cloud_tweet.loc[df_word_cloud_tweet['Date'] == date], 10)
pyLDAvis.display(plot_event)

April 8th, 2022 - Missile strike on railway station

# April 7th, 8th and 9th
prev_day = datetime.date(2022, 4, 7)
day = datetime.date(2022, 4, 8)
next_day = datetime.date(2022, 4, 9)
count_ens_model_prev_day = sentiment_analysis(prev_day, df)
count_ens_model_day = sentiment_analysis(day, df)
count_ens_model_next_day = sentiment_analysis(next_day, df)

# Plotting the pie chat
pie_plot(count_ens_model_prev_day, count_ens_model_day, count_ens_model_next_day, 'April 7th', 'April 8th', 'April 9th')

# Relevant terms
date = next_day.strftime('%Y-%m-%d')
twitter_sentiment_event, plot_event = LDA_Event(df_word_cloud_tweet.loc[df_word_cloud_tweet['Date'] == date], 10)
pyLDAvis.display(plot_event)

May 17th, 2022 - Mariupol defenders surrender to Russia

# May 16th, 17th and 18th
prev_day = datetime.date(2022, 5, 16)
day = datetime.date(2022, 5, 17)
next_day = datetime.date(2022, 5, 18)
count_ens_model_prev_day = sentiment_analysis(prev_day, df)
count_ens_model_day = sentiment_analysis(day, df)
count_ens_model_next_day = sentiment_analysis(next_day, df)

# Plotting the pie chat
pie_plot(count_ens_model_prev_day, count_ens_model_day, count_ens_model_next_day, 'May 16th', 'May 17th', 'May 18th')

# Relevant terms
date = next_day.strftime('%Y-%m-%d')
twitter_sentiment_event, plot_event = LDA_Event(df_word_cloud_tweet.loc[df_word_cloud_tweet['Date'] == date], 10)
pyLDAvis.display(plot_event)

June 30th, 2022 - Ukraine fetes Russian pullback from strategic Snake Island outpost

# June 29th, 30th and July 1st
prev_day = datetime.date(2022, 6, 29)
day = datetime.date(2022, 6, 30)
next_day = datetime.date(2022, 7, 1)
count_ens_model_prev_day = sentiment_analysis(prev_day, df)
count_ens_model_day = sentiment_analysis(day, df)
count_ens_model_next_day = sentiment_analysis(next_day, df)

# Plotting the pie chat
pie_plot(count_ens_model_prev_day, count_ens_model_day, count_ens_model_next_day, 'June 29th', 'June 30th', 'July 1st')

# Relevant terms
date = next_day.strftime('%Y-%m-%d')
twitter_sentiment_event, plot_event = LDA_Event(df_word_cloud_tweet.loc[df_word_cloud_tweet['Date'] == date], 10)
pyLDAvis.display(plot_event)

July 3rd, 2022 - Last major Ukraine-controlled city in Luhansk region falls to Russia

# July 2nd, 3rd and 4th
prev_day = datetime.date(2022, 7, 2)
day = datetime.date(2022, 7, 3)
next_day = datetime.date(2022, 7, 4)
count_ens_model_prev_day = sentiment_analysis(prev_day, df)
count_ens_model_day = sentiment_analysis(day, df)
count_ens_model_next_day = sentiment_analysis(next_day, df)

# Plotting the pie chat
pie_plot(count_ens_model_prev_day, count_ens_model_day, count_ens_model_next_day, 'July 2nd', 'July 3rd', 'July 4th')

# Relevant terms
date = next_day.strftime('%Y-%m-%d')
twitter_sentiment_event, plot_event = LDA_Event(df_word_cloud_tweet.loc[df_word_cloud_tweet['Date'] == date], 10)
pyLDAvis.display(plot_event)

September 12th, 2022 - Ukraine reclaims more territory, reports capturing many POWs

# September 11th, 12th and 13th
prev_day = datetime.date(2022, 9, 11)
day = datetime.date(2022, 9, 12)
next_day = datetime.date(2022, 9, 13)
count_ens_model_prev_day = sentiment_analysis(prev_day, df)
count_ens_model_day = sentiment_analysis(day, df)
count_ens_model_next_day = sentiment_analysis(next_day, df)

# Plotting the pie chat
pie_plot(count_ens_model_prev_day, count_ens_model_day, count_ens_model_next_day, 'September 11th', 'September 12th', 'September 13th')

# Relevant terms
date = next_day.strftime('%Y-%m-%d')
twitter_sentiment_event, plot_event = LDA_Event(df_word_cloud_tweet.loc[df_word_cloud_tweet['Date'] == date], 10)
pyLDAvis.display(plot_event)

October 10th, 2022 - Russia rained cruise missiles on busy Ukrainian cities

# October 9th, 10th and 11th
prev_day = datetime.date(2022, 10, 9)
day = datetime.date(2022, 10, 10)
next_day = datetime.date(2022, 10, 11)
count_ens_model_prev_day = sentiment_analysis(prev_day, df)
count_ens_model_day = sentiment_analysis(day, df)
count_ens_model_next_day = sentiment_analysis(next_day, df)

# Plotting the pie chat
pie_plot(count_ens_model_prev_day, count_ens_model_day, count_ens_model_next_day, 'October 9th', 'October 10th', 'October 11th')

# Relevant terms
date = next_day.strftime('%Y-%m-%d')
twitter_sentiment_event, plot_event = LDA_Event(df_word_cloud_tweet.loc[df_word_cloud_tweet['Date'] == date], 10)
pyLDAvis.display(plot_event)

November 9th, 2022 - Russia says it's withdrawing from the key city of Kherson, but Ukraine is skeptical

# November 8th, 9th and 10th
prev_day = datetime.date(2022, 11, 8)
day = datetime.date(2022, 11, 9)
next_day = datetime.date(2022, 11, 10)
count_ens_model_prev_day = sentiment_analysis(prev_day, df)
count_ens_model_day = sentiment_analysis(day, df)
count_ens_model_next_day = sentiment_analysis(next_day, df)

# Plotting the pie chat
pie_plot(count_ens_model_prev_day, count_ens_model_day, count_ens_model_next_day, 'November 8th', 'November 9th', 'November 10th')

# Relevant terms
date = next_day.strftime('%Y-%m-%d')
twitter_sentiment_event, plot_event = LDA_Event(df_word_cloud_tweet.loc[df_word_cloud_tweet['Date'] == date], 10)
pyLDAvis.display(plot_event)

December 6th, 2022 - Ukraine strikes another Russian air base

# December 5th, 6th and 7th
prev_day = datetime.date(2022, 12, 5)
day = datetime.date(2022, 12, 6)
next_day = datetime.date(2022, 12, 7)
count_ens_model_prev_day = sentiment_analysis(prev_day, df)
count_ens_model_day = sentiment_analysis(day, df)
count_ens_model_next_day = sentiment_analysis(next_day, df)

# Plotting the pie chat
pie_plot(count_ens_model_prev_day, count_ens_model_day, count_ens_model_next_day, 'December 5th', 'December 6th', 'December 7th')

# Relevant terms
date = next_day.strftime('%Y-%m-%d')
twitter_sentiment_event, plot_event = LDA_Event(df_word_cloud_tweet.loc[df_word_cloud_tweet['Date'] == date], 10)
pyLDAvis.display(plot_event)

January 4th, 2023 - Russian military under scrutiny after 89 soldiers were killed in single Ukraine artillery attack

# January 3rd, 4th and 5th
prev_day = datetime.date(2023, 1, 3)
day = datetime.date(2023, 1, 4)
next_day = datetime.date(2023, 1, 5)
count_ens_model_prev_day = sentiment_analysis(prev_day, df)
count_ens_model_day = sentiment_analysis(day, df)
count_ens_model_next_day = sentiment_analysis(next_day, df)

# Plotting the pie chat
pie_plot(count_ens_model_prev_day, count_ens_model_day, count_ens_model_next_day, 'January 3rd', 'January 4th', 'January 5th')

# Relevant terms
date = next_day.strftime('%Y-%m-%d')
twitter_sentiment_event, plot_event = LDA_Event(df_word_cloud_tweet.loc[df_word_cloud_tweet['Date'] == date], 10)
pyLDAvis.display(plot_event)

February 8th, 2023 - Zelenskiy Appeals for Fighter Jets

# February 7th, 8th and 9th
prev_day = datetime.date(2023, 2, 7)
day = datetime.date(2023, 2, 8)
next_day = datetime.date(2023, 2, 9)
count_ens_model_prev_day = sentiment_analysis(prev_day, df)
count_ens_model_day = sentiment_analysis(day, df)
count_ens_model_next_day = sentiment_analysis(next_day, df)

# Plotting the pie chat
pie_plot(count_ens_model_prev_day, count_ens_model_day, count_ens_model_next_day, 'February 7th', 'February 8th', 'February 9th')

# Relevant terms
date = next_day.strftime('%Y-%m-%d')
twitter_sentiment_event, plot_event = LDA_Event(df_word_cloud_tweet.loc[df_word_cloud_tweet['Date'] == date], 10)
pyLDAvis.display(plot_event)

March 7th, 2023 - The video of Ukrainian Soldier getting executed

# March 6th, 7th and 8th
prev_day = datetime.date(2023, 3, 6)
day = datetime.date(2023, 3, 7)
next_day = datetime.date(2023, 3, 8)
count_ens_model_prev_day = sentiment_analysis(prev_day, df)
count_ens_model_day = sentiment_analysis(day, df)
count_ens_model_next_day = sentiment_analysis(next_day, df)

# Plotting the pie chat
pie_plot(count_ens_model_prev_day, count_ens_model_day, count_ens_model_next_day, 'March 6th', 'March 7th', 'March 8th')

# Relevant terms
date = next_day.strftime('%Y-%m-%d')
twitter_sentiment_event, plot_event = LDA_Event(df_word_cloud_tweet.loc[df_word_cloud_tweet['Date'] == date], 10)
pyLDAvis.display(plot_event)

# code to find the longtitude and latitude of locations and use in Tableau
# df_topic_neg["Location"] = df_topic_neg["Location"].astype(str)
# df_topic_pos["Location"] = df_topic_pos["Location"].astype(str)
#location_negative = df_topic_neg["Location"].str.cat(sep=' ')
#location_positive = df_topic_pos["Location"].str.cat(sep=' ')
#import spacy
#nlp = spacy.load('en_core_web_sm')
#doc_neg = nlp(location_negative)
#doc_pos = nlp(location_positive)
# Extract the named entities and create a list of tuples
#entities_neg = [(ent.text, ent.label_) for ent in doc_neg.ents]
#entities_pos = [(ent.text, ent.label_) for ent in doc_pos.ents]
# Create a Pandas DataFrame from the list of tuples
#df1_negative = pd.DataFrame(entities_neg, columns=['entity_text', 'entity_label'])
#df1_positive = pd.DataFrame(entities_pos, columns=['entity_text', 'entity_label'])
#df1_negative["entity_label"].value_counts()
#df1_positive["entity_label"].value_counts()
#df1_negative = df1_negative[(df1_negative["entity_label"] == "GPE") | (df1_negative["entity_label"] == "LOC")]
#df1_positive = df1_positive[(df1_positive["entity_label"] == "GPE") | (df1_positive["entity_label"] == "LOC")]
#location = df1_negative["entity_text"]
#from geopy.geocoders import Nominatim
#geolocator = Nominatim(user_agent="my-app")
#results = []
#for loc in location:
    #location = geolocator.geocode(loc)
    #if location is not None:
        #results.append(( location.latitude, location.longitude))
    #else:
        #results.append(( None, None))
#df_neg = pd.DataFrame(results, columns=['latitude','longitude'])
#location = df1_positive["entity_text"]
#from geopy.geocoders import Nominatim
#geolocator = Nominatim(user_agent="my-app")
#results = []
#for loc in location:
    #location = geolocator.geocode(loc)
    #if location is not None:
        #results.append(( location.latitude, location.longitude))
    #else:
        #results.append(( None, None))
#df_pos = pd.DataFrame(results, columns=['latitude','longitude'])